A data-oriented shopfloor management in the production context: a systematic literature review

‘Data is the new oil’ — Humby used this quote in 2006 to emphasise the fundamental importance of data. Palmer [1] extended the quote with an analogy to crude oil: ‘Data is like crude oil. It’s valuable, but if it’s not refined, it can’t really be used’. This is because data, in its unrefined state, is valuable but has limited use. Over time, new terms such as ‘big data’ and ‘datafication’ have emerged, indicating an increase in data volumes and the evolution of data essence. The concept of datafication, which involves the comprehensive collection, storage and analysis of all data, acts as a catalyst for big data. This recognition has elevated data to a higher stage of development, now recognised as a corporate asset and the basis for innovative business models [2].

Despite the emerging potential associated with increasing data orientation, a company striving for excellence must not lose sight of its fundamental activities. This is the role of daily shopfloor management (SFM), which ensures operational performance while continuously implementing productivity improvements on the ongoing journey towards operational excellence [3].

Combining the two themes of ‘data’ and ‘SFM’, we can see that data is already being used in SFM, even in the status quo. Gisi [3] refers to this as the ‘lifeblood’ of any manufacturing company. On the one hand, relevant key performance indicators (KPIs) can be derived and monitored, such as scrap rates, employee efficiency, throughput times, cycle times or inventories. On the other hand, they provide an essential basis for objective and rational decisions to successfully manage the shopfloor. Deming’s quote ‘Without data, you’re just another person with an opinion’ underlines this aspect [3,4,5]. However, terms such as ‘next generation shopfloor’, ‘digital shopfloor management’ (dSFM) or ‘smart shopfloor management’ are possible evolutionary stages of traditional SFM. Further developments in the area of data are contributing significantly to this leap in development. In this context, it is important to relate key terms such as big data, data analytics, and real-time data processing and analysis to SFM. Terms such as machine learning (ML), artificial intelligence (AI), data mining or predictive/prescriptive analytics sometimes stand for possible technologies and methods for the meaningful use of data and can therefore be used in combination with the key terms [3, 6,7,8].

There are certainly challenges associated with this development. Data collection and processing for the shopfloor still accounts for 57% of the time, the rate of automatically processed machine data is only 5% and only 17.5% of the companies surveyed have a dSFM. However, components of dSFM within the study were not only data, but also digitalisation in general, for example through apps or visualisations [9, 10]. For example, according to Clausen et al. [11], only 7% of dSFM boards are in use. According to a study by the Karlsruhe Institute of Technology (KIT) based on expert interviews, the SFM is largely used as a paper-based or analogue system in the status quo. The benefits for companies in terms of Industry 4.0 maturity are generally limited to visibility and transparency. Established approaches to forecasting and autonomy are therefore not to be found. Information availability is usually dominated by periodic queries. As a result, event-driven approaches are rare [12]. Zhuang et al. [13] highlight this especially for assembly shopfloors.

The conclusion is that data must be given high priority and is essential for long-term business success, including in SFM. In SFM, these are already being used in the status quo, but due to the dynamic and fast-moving developments, there are still challenges in practice. In addition, there are many terms related to data and SFM that are not clearly defined. For this reason, the following research questions (RQs) will be answered through a systematic literature review:

• RQ1: What is a data-oriented SFM?

• RQ2: What is the significance of data in SFM in the manufacturing context?

• RQ3: How should data-oriented SFM be structured?

This also includes clarifying the question of what data-oriented SFM actually means and how it relates to dSFM.

Remark: This article and these research questions focus on discrete manufacturing. This does not exclude the possibility that aspects may be equally applicable to process manufacturing, but due to different frameworks and circumstances on the shopfloor, process manufacturing may differ further.

In principle, two approaches can be used to research specific subject areas: systematic literature reviews, also known as the bibliographic method, or traditional literature reviews. The latter approach is also known as the concentric circle, snowball or avalanche method [14]. The workload and time required for systematic searches is very high compared to the traditional approach [15]. However, there are clear arguments in favour of systematic literature searches: clarity, validity and reproducibility [14, 16]. These predominantly pro arguments are the reason why the systematic or bibliometric approach is used for the following literature review.

2.1 Framework of the literature analysis process

The following systematic literature review procedure is based on the seven-step process described by Fink [16]:

1. 1.

   Select research question(s)

2. 2.

   Select bibliographic database(s)

3. 3.

   Choose search term(s)

4. 4.

   Definition of methodological filter criteria

5. 5.

   Definition of practical filter criteria

6. 6.

   Realisation of the review process

7. 7.

   Synthesis of results

The research questions have already been clarified in Section 1. The choice of bibliographic databases is based on several sources. Web of Science is one of the largest databases in the international environment in terms of importance and quantity/scope. It is also highly relevant in relation to the generic topic of ‘Industry 4.0’ [17, 18]. The selection is complemented by two other databases that are considered to be among the most comprehensive sources. The first is ScienceDirect, one of the leading online databases with an equally high significance for researchers on the topic of Industry 4.0, and the second is WISO, a database with a strong practical focus [17,18,19,20].

The third step is to define search terms that will be used to find relevant publications in the selected databases. These are derived on the basis of the research questions defined at the beginning. In this case, four relevant subject areas can be extracted:

Shopfloor, Management, Data and Production

Based on these subject areas, related keywords and subsequently operationalised search terms are derived. The aim here is to transfer these into a search algorithm that forms the basis for systematic searches within the selected databases. The operationalisation can be implemented using truncation, wildcards or a phrase search, among others [20]. Table 1 provides an overview of the four topic areas, the derived key terms and the operationalised search terms.

A search algorithm based on Boolean operators was then derived from the operationalised search terms. Particularly in view of the very broad range of topics, the approach chosen is to assume the central core topic of Shopfloor with a corresponding truncation in the title. As a result, publications that do not deal with and discuss the shopfloor as a core topic are excluded. The operationalised search terms for the subject areas of Management, Data and Production, on the other hand, are embedded in a so-called topic search. The topic search searches the field tags title, abstract and keywords using OR logic, so that it inherently produces a wider range of search results than the title search. The underlying algorithm for the database search, which can also be seen as a methodological filter criterion, is as follows:

In addition to the methodological filter criteria, the practical filter criteria must be defined. Duplicates and availability/retrievability are defined as practical filter criteria with an exclusive character. One of the practical filter criteria with an inclusive character is the publication language. Only publications available in English are filtered. Secondly, a date filter is used, which only includes sources that are not older than ten years at the time of the query (i.e. publications from 2014 onwards). The date of publication is used as a restriction because, on the one hand, it can be assumed that existing promising contributions have been directly or indirectly cited in the current literature or have been used to build on it. On the other hand, the state of the art in the data context is considered to be so fast-moving that contributions older than ten years have to be classified as partially outdated and therefore in need of revision.

2.2 Realisation of the systematic literature analysis

In order to implement the review process according to process step 6 mentioned in Section 2.1, a multi-stage process flow was designed, which is summarised in Fig. 1. This is based on various existing models and processes, but has been adapted to our own needs with regard to the process flow described in the previous section [21,22,23].

Review process of the systematic literature analysis [21,22,23], modified

Full size image

The search algorithm defined in Section 2.1 forms the basis of the identification step in Fig. 1. As all the selected databases offer the possibility of advanced search based on Boolean operators, but the selection options and descriptions of the field tags differ between the databases and there are some technical limitations in the search, the individual search algorithms for each database are documented in Table 8.

The number of results, after removing duplicates and categorised by language and year of publication, is shown in Fig. 2.

Results after applying the search algorithms in the bibliographic databases

Full size image

As can be seen in Fig. 2, although there are various publications prior to the period considered relevant from 2014 onwards, the number of publications has increased in the last decade in particular, due to the increasing importance and attention given to the topics of big data and datafication, as mentioned in Section 1. Specifically, of the total of 437 results after removing duplicates, around two-thirds (63.4%) relate to publication years greater than or equal to 2014.

After excluding further documents with a publication date older than 2014 (n = 160), document languages other than English (n = 70) and missing access options, 211 articles remained. The titles and abstracts of all remaining publications were then read in order to make a selection based on the relevance of the content and the fulfilment of scientific quality requirements. In order to objectify the aforementioned relevance assessment procedure, each publication was assigned a priority number. The priority numbers, which ranged from one to ten, were categorised into five classes and described qualitatively (see Table 2).

Those publications that could already be clearly assigned to the first class between one and two points on the basis of the title and abstract were categorised as ‘not relevant’ at this stage and thus discarded from further consideration (n = 109). The remaining 100 publications, after exclusion of non-peer-reviewed articles (n = 2), were read in full and categorised qualitatively on the basis of the full text, taking into account the priority number. This included 21 publications that were again assigned to the first priority category with one to two points, but only after the full text had been evaluated. As in the previous step, these were classified as ‘not relevant’ and therefore excluded from the next steps. The remaining 79 publications with a score of 3 or more points form the basis of the following literature analysis.

2.3 Literature review

The synthesis of the results as the final process step 7 mentioned in Section 2.1 is further explained in Sects. 3–5 and in concluding Section 6. In order to better categorise and structure the synthesis of results, the included publications were then evaluated at three levels of consideration: Technology, Organisation and People. This framework was chosen because the factors are suitable for conceptual analysis, for explaining adoption/change, for examining interrelationships and for successful technology implementation, for example in the field of information technology (IT) [24]. As the topic of data is linked to IT, and the aspects mentioned (e.g. relationships between the dimensions, change in SFM and conceptual considerations) play an important role in the context of data-oriented SFM, these three dimensions are considered and evaluated separately below. This should not only contribute to a structured analysis of the literature, but also provide a basis for future conceptual developments.

In addition to their relevance and priority according to Table 2, the quality of the content was also assessed in terms of any focal points and conceptual research gaps. Accordingly, the following chapters are organised according to these three levels of consideration and the results of the respective key topics are described in the respective chapters. Each publication was scored by Harvey balls on the extent to which each level of consideration was met. The scoring scheme is shown in Table 3.

The aim of applying the scoring scheme in Table 3 to each article was to identify where the focus or research gaps in the existing literature lie at each level of consideration, from a conceptual point of view, in relation to the topic at hand. As a result, all 79 publications were assessed in terms of technology, people and organisation, resulting in three Harvey balls per publication. Based on this, Table 4 was used to create a matrix showing how many scores (Harvey balls) were awarded within each level of consideration. The higher the number of articles in an increasing level of consideration (red = low number; green = high number), the greater the depth of discussion of the level of consideration in the literature. For example, many publications omit the levels of consideration ‘people’ (n = 37) and ‘organisation’ (n = 45).

In addition to the scoring from Table 3, each publication was assigned one or more tags. The aim of the tags was to characterise the content discussed in each publication and to detail the levels of consideration. This made it possible to organise the literature thematically in order to derive the table of contents for Sects. 3–5. Unlike scoring, however, tagging was an iterative process within the literature analysis, as the tags were created gradually based on the content focus of the publications. For example, the area of ‘technology’ was divided into more thematic focus areas/tags than ‘people’ or ‘organisation’ because the literature discussed this level of consideration and the underlying subject areas more intensively. New tags were assigned when a publication could not be categorised under existing tags. In addition, certain tags were grouped together in final iteration loops if they covered similar topics and had a high degree of overlap. Table 5 shows the result of the content focus (red = very little focus in the literature; green = intensively discussed) from the analysed literature.

A combined quantitative interpretation of Table 4 and Table 5 already shows that, compared to people and organisation, technology in particular is relatively over-represented in the literature. This applies not only to the content of the tags (Table 5), but also with regard to the qualitative categorisation of holistic concepts with a comprehensive character (Table 4). If we look at the publications which, according to Table 4, are rated as having the best possible degree of fulfilment per level of consideration (= Harvey ball completely filled), at least one of the other two levels of consideration is only partially fulfilled (= Harvey ball only quarter or half-filled). It can therefore be concluded that there is no all-encompassing concept that takes full account of the three levels of consideration, including their interdependencies, synergies and overlaps.

In the literature, the field of technology is discussed from different perspectives. These perspectives can be divided into specific applications on the one hand and technical frameworks on the other, which are described below. Finally, the level of technological maturity, which combines both perspectives, will be discussed.

3.1 Results: technical applications

The following section discusses use cases where data is a key contributor to adding value on the shopfloor. These range from maintenance to quality and production planning. As these applications are either named as concrete decision-making approaches or indirectly contribute to data-based decision-making on the shopfloor, they can also be summarised under this keyword and the intention ‘decision-making’.

3.1.1 KPIs/monitoring

A key use case for data-oriented decision-making is the calculation of KPIs or general shopfloor monitoring and performance management. These can be used to create data transparency, monitor production and provide a basis for decision-making when taking action. In a data-related context, the most automatic and accurate collection and the best possible accessibility (e.g. remote or real-time) represent significant advantages in terms of efficiency and effectiveness. Alternatively, analogue SFM boards are also possible [11, 25, 26]. The metric overall equipment effectiveness (OEE) is an example that provides clear benefits for management and workers [27]. It is also possible to analyse worker or machine performance in order to calculate and monitor KPIs such as lead time, lasting time, finishing time or general efficiency [28,29,30,31]. In addition to specialised KPIs for machine or personnel performance, real-time monitoring of energy efficiency or waste reduction is also possible [32,33,34].

3.1.2 Process and design planning

Process and design planning is important both for optimising production processes on the shopfloor and for the overall product design. This means that, in the best case, an optimal product and process design should be in place before the design is finalised. Data plays a key role here, as simulations can be run on the shopfloor using a digital image of product and process knowledge. Based on this, optimised decision-making with respect to possible scenarios and iterative product and process improvements can be realised prior to actual implementation on the shopfloor [35, 36]. The simulations consider not only cost or efficiency, but also, for example, effectiveness in terms of product quality and machine accuracy through optimal machine parameters or minimisation of tool wear [37]. The digital twin is the keyword here, but this is explained in more detail in Section 3.2.2.

3.1.3 Maintenance

Maintenance in the context of SFM is discussed not only from a maintenance cost perspective, but also because machine failures can lead to unplanned supply bottlenecks in subsequent processes or customer deliveries. There are already several approaches in the literature on how to perform preventive maintenance based on real-time monitoring. Mourtzis et al. [38] present a cloud-based approach for condition-based preventive maintenance (CBPM), which uses sensor data from the machine and input from the machine tool operator (i.e. busy, available, down) to report condition information in near real-time on mobile devices without the need to be physically present on the shopfloor. Predictive and prescriptive maintenance using ML are further worth mentioning and important developments discussed in the literature. However, due to the size of the subject, these are not described in detail here.

3.1.4 Scheduling

In more advanced approaches, the results of condition-based maintenance are used as the basis for the associated adaptive scheduling. If a disturbance is detected on the shopfloor that affects the scheduled tasks, the rescheduling process is triggered. Alternatively, a query-based approach can be chosen, in which rescheduling takes place according to defined frequencies or a manual trigger. Mourtzis and Vlachou [39] use the disruption-based approach, where a disruption can be caused either by an unplanned machine state (e.g. downtime) or by a delay in the planned processing time. Rescheduling becomes necessary as soon as the makespan is significantly affected by disturbances (e.g. influence on post-conditioning tasks). The monitoring data can then be used to apply a multi-criteria decision algorithm to implement the rescheduling, which has a positive impact on efficiency by reducing the manual scheduling and monitoring workload. If a machine cannot provide all the necessary data directly, radio frequency identification (RFID) technology can help to smartify objects in production. For example, Zhong and Xu [40] or Zhang et al. [41] describe such an approach in the context of production planning and scheduling on the shopfloor, using automatic identification (auto-ID) devices to collect real-time data from manufacturing resources. Indoor positioning systems (IPS) can also be used as part of a scalable and reliable real-time location system (RTLS), for example to compare pre-calculated cycle times, utilisation rates or line balancing with reality using statistical approaches [42]. In some cases, further downstream agents within a multi-agent model help to effectively use the collected data to realise the best possible production plan, for example by taking into account machine capabilities and monitoring processes. Knowledge management–based approaches are also possible [43, 44]. The importance of consistency and interaction between different data sources and systems such as enterprise resource planning (ERP), product lifecycle management (PLM) or manufacturing execution system (MES), as well as communication and decision autonomy, should also be emphasised for dynamic scheduling [36, 45,46,47].

3.1.5 Shopfloor logistics

In addition to the production planning and scheduling use case just discussed, production/material logistics also plays an important role on the shopfloor and forms part of the basis for the data-based and autonomous scheduling mentioned in Section 3.1.4. It is the basis for well-functioning, effective and efficient production, as delays in the arrival of materials, for example, have a major impact on the production schedule. Auto-ID (i.e. RFID) is a possible and important technology for the smartification of logistics objects in this context. For example, material-related logistics data on the shopfloor can be used to monitor and optimise bottlenecks in shopfloor logistics, performance deviations, material stock/work in progress (WIP) levels or on-time delivery of shopfloor logistics [48,49,50]. The associated decisions can also be realised in dynamic and autonomous systems, as in the applications mentioned above. For example, in a method of Wang et al. [51], a proactive material handling approach based on predictions of various future states (e.g. manufacturing system prediction, trolley status prediction, WIP processing time prediction, logistics task prediction) is used to allocate tasks in shopfloor logistics in an intelligent and optimised way.

3.1.6 Combined approaches

In addition to dedicated algorithms for the applications mentioned, there are also combined approaches in the literature. In this paper, this is understood as the idea that the intended application uses multiple use cases for decision-making (e.g. quality and maintenance are included in scheduling) or that the solution supports multiple use cases simultaneously (e.g. generic tool). This breaks down silos and promotes more holistic decisions. For example, Kumari and Kulkarni [52] present a simulative and predictive approach in which decision-making is based on a look-ahead mechanism quantified with different impact factors, taking into account Shannon’s entropy. Here, the static complexity in a system (S) is calculated according to Eq. (1), taking into account the number of resources (R), the probability (p_ij) of the resource (j) being in state (i) and the number of possible states (N_j) of a resource (j).

The complexity for a machine (M) during the time duration (T) is described in Eq. (2). The total number of jobs (n) assigned to a machine within the time period is divided into six states (s). These are made up of a quality state (good, bad) and a time state (early, on time, late), so that their combination gives six states that can potentially be assigned to a job.

In addition to the complexity, a penalty (Eq. 3) is also quantified, which consists of a schedule penalty (P_SC) and a quality penalty (P_q). Any deviation from ‘On Time-Good’ for P_SC and any deviation from ‘Bad’ for P_q results in a penalty, whereby the priority of a job is included with a positive multiplier.

However, the simulation also takes into account 11 uncertainties that affect the job schedule, quality and penalty (e.g. batch size and priority change, setup or process failure, probability of due date change, time variability for corrective maintenance, or maintenance staffing). The combination of penalty level and complexity level, categorised according to the knowledge and thresholds of the decision maker, forms the Undesirability Index and the associated decision rules shown in Table 6.

This approach is an example of how quantification-based shopfloor decision-making has been discussed in the existing literature. On the one hand, any holistic approach based on numerical data facts reduces the degree of subjectivity for decision makers. On the other hand, the effectiveness of decisions is increased because a pre-defined set of rules is simultaneously followed within the simulation and prediction and clear instructions are given to the decision maker (e.g. immediate intervention is necessary due to increased penalty if a rush order has to be processed but the machine is in non-critical maintenance).

AI is playing an increasingly important role on the shopfloor for such a combined approach, but also for the technical applications mentioned in Sections 3.1.1–3.1.5. ML as a subset of AI is particularly interesting for SFM in the context of data, as ML can adaptively gain experience and knowledge from collected data. For example, it can identify characteristics and relationships between variables, build complex statistical models, perform fault diagnosis and predictions, and enable predictive or prescriptive decision-making on the shopfloor [53, 54]. There are many ML algorithms that can be used for this, such as regression, decision trees, support vector machines and neural networks in various forms [55]. In addition to use cases that focus on information flow, combined ML applications are also conceivable that relate to material flow and indirectly contribute to information-based applications such as those in Sections 3.1.1–3.1.5 by using ML algorithms. Examples are intelligent ML-based visual quality control systems or ML-based routing of automated guided vehicles [56, 57].

Furthermore, there are combined approaches that support decision-making on a generalised basis and with qualitative statements. For example, there are expert systems or knowledge management systems based on generic decision trees or recommender engines to help shopfloor workers make decisions and solve problems. The workers can communicate with the tool via a dialogue module in the form of an interview and are guided to initial solution ideas at a very abstract level [58, 59]. The externalisation of tacit knowledge plays an important role in such approaches [60]. AI is also becoming increasingly important in processing linguistic and cognitive content and interacting with people on the shopfloor. For example, natural language processing (NLP) is an important area where human language is processed with the aim of understanding and, in this case, using it profitably on the shopfloor. Possible use cases include voice assistants, chatbots and assisted knowledge management [61]. Among other things, large language models (LLM) are a technical option for implementing NLP on the shopfloor [62].

Search functions for previously solved problems, tracking and forecasting of solution progress, reminders for next steps, and automated escalations are further examples of generic support for systematic problem solving that can be integrated into the above applications or used as a combined solution [9]. In addition, there are already case studies for autonomous networks and the associated fully autonomous decision-making [63].

3.1.7 Conclusion & Gaps

Looking at the individual applications, the implementation of e.g. scheduling, maintenance-based approaches or other (even autonomous) decision-making use cases are well researched in the literature and confirmed by case studies. The dedicated use of data in SFM brings a number of benefits, such as increased efficiency (e.g. through faster or automated execution of tasks/decisions), reduced costs (e.g. in maintenance), increased effectiveness (e.g. through better decisions) or a general increase in output or performance on the shopfloor. The integration of AI/ML is also becoming increasingly important in order to further optimise the efficiency and effectiveness of the data-oriented applications. However, it can be summarised that there is no holistic consideration of a possible heterogeneous production/shopfloor in terms of manufacturing processes (e.g. production with individual machines, linked machines and manual production) in connection with heterogeneous data availability and different interlocking use cases as mentioned above. On the one hand, many of the mentioned approaches focus on machine data, which seems to be ‘easier’ to provide in real-time compared to, for example, manual assembly. Zhuang et al. [13] emphasise this aspect by pointing out that reactive approaches with manual data work or paperwork are still often used, especially in assembly shopfloors. On the other hand, concepts for capturing manual activities (e.g. with auto-ID) were also mentioned, but their holistic consideration in a heterogeneous shopfloor was not taken into account. When heterogeneous manufacturing and data environments were considered, there was a particular focus on use cases and technologies, such as Ren et al. [48] on material delivery in combination with auto-ID, instead of discussing holistic concepts for the shopfloor. In addition, many approaches are based on individual work centres (e.g. in the form of machines) or functional aspects (e.g. material logistics or production planning) on the shopfloor and are not evaluated from a (possibly heterogeneous) value stream/process perspective. Although some of the combined approaches mentioned in Section 3.1.6 above take into account other conditions, such as upstream or downstream processes, it is not entirely clear how these interact with each other. Mohammed et al. [64] confirmed this aspect in a literature review, stating that although goal-oriented performance measurement is the present and future generation in SFM, process-centric/value stream-oriented SFM guidelines in terms of KPIs are lacking.

3.2 Results: technical environment

In order to realise these applications, there are requirements for the technical environment on the shopfloor and in production. The approaches evaluated in the literature are discussed below.

3.2.1 Data management environment

In the applications described above (maintenance, scheduling, decision support, etc.), cloud technology is mostly used within the data management infrastructure because it allows ubiquitous access to data by multiple IT tools and users, supports collaboration and overcomes the challenges of collaboration, reconfigurability and adaptability of manufacturing systems to dynamic change [27, 39, 65,66,67]. Therefore, there is widespread agreement in the recent literature that the use of cloud technology makes sense in the context of data in SFM, partly due to these advantages. Nevertheless, it should be noted that hybrid systems with cloud-edge technologies are also used in some cases in production, because edge nodes are more advantageous, for example, for time-critical decisions (low latency) and processing of raw data (proximity to the source) [32]. In the cloud computing paradigm, elastic computing is often seen as a key element. This is particularly necessary when processing large amounts of data, which can occur in the shopfloor, as elastic computing automatically scales computing resources up (provisioning) and down (deprovisioning). This means that the resources available when using elastic computing correspond to the current demand at a given point in time [67].

Based on the literature, a cloud-based or cloud-edge-based ecosystem on the shopfloor can be divided into two layers: the physical shopfloor and the cyber shopfloor. The cyber shopfloor is where cloud or cloud-edge technologies are mainly used [39, 67]. Alternatively, other distinctions are also conceivable (e.g. in machine layer, operational technology (OT) layer, mirroring layer, IT layer, view layer) [68], but this distinction is favoured in the following due to the present shopfloor context. Furthermore, the distinction between the two layers (‘physical’ and ‘cyber’) is in line with the concept of cyber physical systems (CPS), which is directly related to Industry 4.0 in the manufacturing context. A CPS describes the collection and transmission of data from physical components, systems and processes in cyberspace for analysis, monitoring and decision-making. Mathematically, a CPS represents the function of sense, connect, content and share [27]. In addition to CPS, cyber-physical production systems (CPPS) are associated with the manufacturing industry and use the industrial internet of things (IIoT), a network of highly connected and intelligent industrial components, to integrate different CPS nodes. From the point of view of the literature, there is already very good work on the technical realisation and integration of CPS/CPPS/IIoT in production systems [32]. A ‘cyber-physical shopfloor’ (CPSF) is an alternative formulation that represents a similar understanding [69].

After analysing and summarising the literature on shopfloor data management, this paper has been able to derive and distinguish two self-described approaches, which are explained below: a ‘data push system’ and a ‘data pull system’ from the perspective of the physical shopfloor. The push system generates data on the physical shopfloor and makes it available to the cyber shopfloor in real-time wherever possible. In the pull system, the physical shopfloor actively requests data from the cyber shopfloor and uses it, for example, for operational activities. There is no strict dichotomy between these two systems; they can co-exist, interact or even be based on the same technological setup.

‘Data push’ CPPS

As there are many ways to collect data, such as directly from machines, processes, data acquisition devices (DAQ), additional or separate sensors and RFID tags, the data sources in the physical shopfloor can be deductively grouped as data sources (DS). Between the physical and cyber parts, one or more interfaces are required to transfer data (e.g. in terms of interoperability, compatibility and portability). The intention is usually to collect, transmit and provide data in real-time. The collected data is available in the cyber shopfloor, which, as described, is often based on cloud or cloud-edge solutions. Various approaches to structuring and exploiting this space can be found in the literature [39]. For example, as practically described by Terrazas et al. [67], several layers are conceivable within the cyber shopfloor to store and retrieve persistent manufacturing data (knowledge), extract systematic knowledge (analytics), execute back-end logic and analytics orchestration (application), and present actionable information, including managing external interactions with end users (presentation). In addition, architectures such as the concept of Li et al. [70] already exist for challenging technical requirements such as heterogeneous data management, networking of resources and handling of domain knowledge. A condensed summary visualisation of the data push CPPS using these exemplary layers is shown in Fig. 3 [27, 39, 45, 67, 71,72,73]. In this context, however, it should be noted that there is no universally valid design for a CPPS, but many different perspectives and architectures [65]. Therefore, this simplified illustration (also for the following subsections) is suitable to represent the basic structure of a CPPS.

Data push CPPS

Full size image

‘Data pull’ CPPS

In contrast to a data push CPPS, which plays an important role in e.g. analysis or decision-making, a data pull CPPS, as terminologically defined in this paper, is used for operational execution on the shopfloor. For example, Li et al. [66] present a cloud-based shopfloor control system (SFCS), which in the context of this paper is a pull-based CPPS. The intention here is to manage all the data required for operations on a cloud-based basis and thus supply all workstations centrally. The aim is to eliminate client-based systems so that, for example, the provision of new production versions or production documents, the integration of new products or product changes, and changes in production planning or software/system maintenance are not carried out on a workstation-specific basis, but can be accessed centrally from a single source of truth in the cloud. This means that the last instance in the physical shopfloor does not fall back on local data when information is required, but requests it from the cyber shopfloor using the pull principle, as illustrated in Fig. 4. In addition to Fig. 3, Fig. 4 shows examples of possible source systems such as an ERP and PLM system, MES, separate databases or manual input are outlined, which, in turn, form the basis for the cyber shopfloor. The data stored in these systems includes anything that can be considered relevant to the shopfloor, such as product data, process data or production data.

Data pull CPPS

Full size image

In addition to the general arguments in favour of using cloud technology, the cloud-based pull approach in particular offers significant efficiency gains (e.g. reduced deployment effort for version changes), a reduction in error rates (e.g. avoiding the use of incorrect versions or labels), productivity gains (e.g. fewer line stoppages due to faster deployments), an improvement in the data environment (e.g. centralised programme management) and the provision of an effective reporting service (e.g. automatic generation of reports with collection of correct versions) [66].

Customer-oriented

CPPS The two concepts outlined in the previous sections (‘Data push’ CPPS and ‘Data pull’ CPPS) focus on the internal physical and cyber shopfloor. However, for reasons of transparency, the need for data availability to external customers is increasing. For example, Ding et al. [32] have discussed the concept of a cyber-physical production monitoring service system (CPPMSS), which is a combination of new information technologies (i.e. CPS/IIoT) and advanced business models such as product service systems (PSS)/production service systems (PnSS). A PSS/PnSS combines a product/equipment with a corresponding service, which is also increasingly used in the manufacturing industry as a business model in connection with equipment providers (i.e. manufacturers pay for the services and functionalities individually, not for the equipment as fixed capital). In this concept, there is a third layer in addition to the physical and cyberspace: the service space. This is where, for example, the manufacturer, the equipment provider and the customer interact. The customer could, for example, make enquiries about production progress or current environmental data (e.g. real-time energy consumption or energy efficiency). An external equipment provider, on the other hand, could, among other things, query equipment statuses (e.g. for maintenance purposes) [32]. Using the earlier terminology of push/pull logic, there is both a push and a pull principle between the cyber layer and the service layer, as data is both requested and, under certain circumstances, fed back (e.g. feedback loops with subsequent actions). Figure 5 illustrates the three-layer framework of the CPPMSS according to Ding et al. [32] based on a cloud-edge technology.

Three-layer framework of CPPMSS [32]

Full size image

Conclusion & Gaps

It can be said that the topic of data management environments and infrastructures has been very well researched in the literature using different approaches and validated with practical case studies. There are various multi-layered concepts with potential solutions and (open source) standards (e.g. REST API, Apache Kafka, UPC-UA) that allow the storage and processing of real-time data — also in relation to the shopfloor. However, as Fig. 3, Fig. 4 and Fig. 5 are very condensed (e.g. not taking into account possible heterogeneity of existing infrastructures, ranging from old machines to a wide variety of interfaces), limitations need to be highlighted. These include low latency and high scalability requirements, as well as the integration of data centres, legacy infrastructure, edge computing devices, on-premises technologies and applications/use cases [13, 45, 74, 75]. At the same time, the increasingly important issue of cyber security needs to be addressed, as it is sometimes not included in the literature [32, 67, 71]. Despite initial attempts to find solutions, these aspects continue to pose challenges.

3.2.2 Digital twin

The digital twin (DT) also plays an important role in the discussion of a CPPS setup and the use cases presented in Section 3.1. DT is a core technology within a CPPS for describing physical entities using digital technologies. Digital representation, synchronous mapping (mirroring the real world), simulation and prediction, and virtual and physical fusion are the four steps in the implementation process of a DT. In the context of the shopfloor, the shopfloor digital twin (SDT) is a description and mapping of all elements and processes of the physical shopfloor. The digital twin shopfloor (DTS) can be understood as a new paradigm of future shopfloor instances of a CPPS/CPS in the production system. It achieves the fusion between physical and cyber shopfloor and even a possible service space through bi-directional interaction [13, 76, 77]. The contents of the all-encompassing DTS are summarised in Fig. 6 [78].

Structure of a digital twin shopfloor (DTS) [78], modified

Full size image

As Fig. 6 only shows the components of the DTS, it should be noted that these, in turn, are bi-directionally connected (meaning of the ‘Connection’ block). The shopfloor digital twin data (SDTD) component brings together data from the physical shopfloor, the SDT and the shopfloor service system. Similarly, the latter three components also interact with each other, e.g. in the form of iterative optimisations, feedback loops or status updates to ensure consistency [75]. In contrast to the data environment of a CPPS described in the previous section, with the SDT within the DTS, not only is data exchanged within the physical and cyber shopfloor, but the physical shopfloor with all its characteristics is also represented virtually. However, DTS/SDT cannot be distinguished from CPPS, but should rather be seen as its core component and core technology. For example, Zhuang et al. [78] have constructed and applied a four-layer SDT framework that makes state predictions based on Markov chains. Figure 7 shows a simplified representation of a possible SDT framework based on this four-layer approach, embedded in the context of a CPPS, taking into account Section 3.2.1 [13, 36, 73]. This is a more detailed view in comparison to Section 3.2.1 and should therefore be regarded as a subset of it. As with CPPS, there are many different models and perspectives, so Fig. 7 summarises and illustrates them in a simplified way. It is important to emphasise that both physical and cyber elements are comprehensively encompassed within the SDT framework, as this is not always the case with some approaches according to a comprehensive literature review by Corallo et al. [76] on the subject of SDT. This is the reason why Corallo et al. [76] have also developed their own holistic concept called hexadimensional shopfloor digital twin (HexaSFDT). Other SDT frameworks can also be found in the literature, such as different seven-level layer models by Yang et al. [73] or Sun et al. [36] who, for example, present PLM systems as part of a management layer. However, any source systems (e.g. PLM, ERP, MES) are to be seen as part of the shopfloor service system and, consequently, of the SDTD according to Fig. 6, so that these data are, in turn, connected bi-directionally with the SDT, as mentioned [13, 75, 79]. It can be concluded from this that the various architectural frameworks and metamodels often follow similar approaches in terms of content, but that the structure and organisation of the various layers diverge.

Exemplary SDT framework in the context of a CPPS

Full size image

Application-specific DTs are also used on the application side, such as in the proactive material handling concept of Wang et al. [51] to make decisions based on an image represented in the virtual environment. Individual DTs can also be seen as an enabler for the production scheduling, shopfloor monitoring and process planning described in Section 3.1, as the various source systems interact with each other in real-time in an integrated framework [36]. In addition, there are already DTS setups for complex (e.g. multi-batch and multi-variety) shopfloors including online predictions based on real-time data and the associated rapid decision-making [41]. In particular, AI is a technology that is playing an increasingly important role in the realisation of use cases within DTS [54]. This is mainly because decision-making use cases require real-time interaction, but this requirement is often a core problem of a DTS. Predictive AI approaches can be used, for example, to compensate for time or content inconsistencies between the physical and cyber shopfloor to continue to ensure effective decision-making. AI is therefore an enabler of synchronisation between state perception and control execution [53]. Consequently, in the present context of DTS, AI helps to apply the use cases described in Section 3.1. This is because with an incorrect database, e.g. due to a time offset between the physical and cyber shopfloor and thus an inconsistent DTS, the real-time possibly AI/ML-supported applications from Section 3.1 would, in the worst case, lead to incorrect decisions.

Conclusion & Gaps:

The structures and possible frameworks of DTSs/SDTs are well described in the literature — even for complex shopfloors. Application-specific DTs are also discussed. These can be seen as enablers for a comprehensive CPPS and the respective applications/use cases. However, some gaps are seen in the fact that the connection and integration of multi-layered use cases such as decision-making for various problems (see examples in Section 3.1), DT-based production optimisation and the associated control dimensioning are not consistently taken into account. Only through this holistic integration and the mapping of more complex SDT-based multi-scenario use cases, including more accurate forecasts and better simulation results, is it possible to provide managers with a faster and more accurate basis for optimised decisions [36, 37, 46, 51, 78]. Corallo et al. [76], for example, confirm these research gaps in a comprehensive literature review on SDT, adding that despite the existence of holistic frameworks, there is a lack of commonly accepted methodologies, including implementation roadmaps and reference standards. Finally, on the subject of performance gaps, it should be noted that AI can partially compensate for these in a DTS (e.g. problems with latency times or real-time requirements), but there is still a need for research in this area in order to ensure the reliability and robustness of an AI-assisted DTS [53].

3.2.3 IoT/IIoT enabling by wireless technologies

If DT/SDT is seen as a subset of a holistic CPPS, then wireless technologies, as a main IoT/IIoT technology, can be seen as a subset of both. Unlike CPPS/DTS, wireless technologies are not so much holistic frameworks as they are dedicated components of the physical shopfloor. These are considered separately below, as they are playing an increasingly important role as a source of real-time data, particularly for the collection of data on the physical shopfloor. Various technologies are possible, such as auto-ID (including e.g. RFID, barcode or voice/face recognition), Bluetooth, Wireless Fidelity (Wi-Fi), Global System for Mobile Communications (GSM), 5G or infrared. The use of such technologies is sometimes referred to as ubiquitous manufacturing (UM). RFID, in particular, is playing an increasingly important role in tracking and tracing production items and collecting real-time production data [29, 80, 81], especially in areas such as the assembly shopfloor, where a lot of manual work is done and automatic machine data provision is not possible [13]. The cloud technology described in Section 3.2 also plays an important role in this context in order to be able to provide the information collected by wireless technologies in real-time [80]. The following applications discussed in the literature can be realised using auto-ID, for example:

• Worker’s performance evaluation [29, 82]

• General real-time shopfloor KPI monitoring (e.g. WIP management, time monitoring, quality) or specific shopfloor logistics like material delivery [13, 28, 48,49,50]

• Production planning and scheduling [40, 81]

If we take the widespread RFID technology as an example, interaction is always required between an object with an RFID tag and an object with an RFID reader, both of which are referred to as a visualised object (VO). For example, RFID readers can be attached to static objects such as workstations, forklifts or machines. Tags, on the other hand, can be attached to boxes, trays or parts [28]. In contrast to RFID, some IPS technologies, such as ultra-wideband (UWB) or radar-based tracking, have the advantage that, in some cases, no operator interaction or attention is required [42]. There are also niche use cases for wireless technologies, such as the collection of health data (e.g. posture data) to promote the well-being and safety of workers on the shopfloor [83].

Conclusion & Gaps:

The functionality and possibility of object smartification and real-time data provision is basically given with wireless technologies and has been well researched in the literature. They can act as an enabler to capture heterogeneous process data and make it available as part of a CPPS architecture [70]. Looking at the research gaps and perspectives, there is often talk of increasing the accuracy and speed of data provision (real-time) and embedding it in real-time decision-making processes. This means optimising the technology and the associated use of data [42, 48, 81, 84]. Nevertheless, it should be noted that these technologies are often ‘only’ an infrastructural means to an end, in order to record missing object data (e.g. due to manual processes) and make it available for intelligent and autonomous decisions.

3.3 Results: technical maturity level

With reference to technical applications and technical environments, different levels of technical maturity of a shopfloor are discussed in the literature. For example, Zhuang et al. [13] have formulated four stages, which are visualised in Fig. 8. There are other evolutionary perspectives in the literature that depict similar stages, but diverge slightly in terms of content and focus [64, 75, 85, 86]. The model by Zhuang et al. [13] should therefore be regarded as exemplary.

Technical shopfloor maturity [13]

Full size image

The first stage is passive/reactive management and control, where most data is collected by paper or manual input. The smaller amounts of data are stored and managed using simple means. Processing is based on historical data and is always delayed. The second stage is extended to include IoT and thus the wireless devices described in Section 3.2.3, such as RFID. This enables real-time data collection, processing, analysis and decision-making. Stage 3 enables foresight value through predictive approaches and thus the fight against uncertainty. This is based on an extended CPPS including digital twins as described in Sects. 3.2.1 and 3.2.2. The final stage 4 changes less in the technical environment with respect to CPPS/DTS, but rather in the final extension of the applications according to Section 3.1. Autonomous decision-making based on prescriptive analysis, including self-adaptation and self-reconfiguration functions within a closed-loop control loop with the CPPS, is seen as the final stage of expansion [13]. In addition to staged maturity levels, there are also concrete target states of a dSFM, in which possible end states with regard to digitalisation on the shopfloor were described on the basis of surveys or questionnaires [87, 88]. The same applies to specific components of the dSFM, such as dSFM boards [11, 25].

Conclusion & Gaps:

Both the stages of expansion of the technical infrastructure and possible use cases for the shopfloor are discussed in the literature in terms of different possible maturity levels. For example, Zhuang et al. [13] use a framework and a first case study to show that it is possible to extend to the fourth level for assembly shopfloors with more challenging requirements. Although this paper is dedicated to the very specific satellite industry, they envisage extending the concept to other complex assembly shopfloors and optimising it as a research perspective. However, these approaches, or other target-oriented approaches such as those of Meissner et al. [87] or Torres et al. [88] did not take into account the shopfloor environment and focused mainly on the application/use case side. In other words, for which type of shopfloor (e.g. in terms of complexity; heterogeneity of production in terms of machinery vs. manual assembly; number and expertise of employees; type of applications or use cases) which maturity level actually appears to make sense. Considering the example of Zhuang et al. a lot of IoT and other infrastructures was retrofitted until a sufficiently established digital twin was established to implement the desired use cases. The topic of cost–benefit will be discussed in more detail in Section 4.1, but it is worth highlighting this gap here. In addition to technical maturity, organisational and human aspects (e.g. leadership and employee acceptance) play an essential role in the implementation of technical improvements [85, 86, 89]. Despite this, these observation levels are under-represented in the technical maturity assessment and are partially discussed in separate social maturity models (see details in the following chapters).

In addition to the technical aspects, the organisational aspects are equally important, as data on the shopfloor can sometimes influence work organisation, activity structures, tasks or decisions on the shopfloor.

4.1 Cost–benefit evaluation

Cost–benefit considerations play a crucial role in the decisions made by an organisation or company. Regardless of the use cases discussed in Section 3, companies often do not even have the technical environment of a CPPS to implement them. This means that investment must first be made in the necessary equipment to create a ‘CPPS-ready’ (or ‘CPS-ready’) shopfloor. Particularly for companies that are not yet CPPS-ready, the question of the benefits and cost-effectiveness of implementation quickly arises [90]. This also applies to individual use cases such as dSFM boards [11]. However, the complexity and interdependencies of processes often make it impossible to predict the outcomes, business value and therefore return on investment of data exploitation. This can lead to a reluctance on the part of management to expand the shopfloor data strategy [45]. In order to objectify this issue, Romero-Silva and Hernández-López [69] classified the shopfloor into different environments based on different characteristics (Table 7), using the exemplary use case of production planning (see Section 3.1.4).

It can be stated that the use case of optimised or autonomous shopfloor scheduling creates competitive advantages mainly in the ‘stress shopfloor’, as this category has, among other things, a high dependency on a constant updating of the planning, different possible routings/configurations and demanding customer requirements (e.g. exact delivery dates and short delivery times). However, in the other shopfloor environments shown in Table 7, it generates limited or no added value. This was investigated experimentally/simulatively using the implementation level (zero, partial, full) on the basis of CPS characteristics (automation, supervision, control and interconnectivity). The quintessence of the hypotheses is that an increase in the CPS/CPPS implementation level does not generally lead to significant improvements and is dependent on the shopfloor environment [69].

In addition, there are application-specific approaches that perform cost–benefit assessments. For example, Lee et al. [35] have developed a framework in which process and design planning decisions (see Section 3.1.2) are evaluated on four levels and operationalised with KPIs. This is intended to contribute to objective decision-making in the planning and design process, always keeping in mind possible outcomes and business perspectives.

These examples show that there is no general correlation between higher CPPS maturity and higher cost-effectiveness. This means that there must be cost–benefit considerations for further developing the CPPS or introducing new technical applications to ensure cost-effectiveness (e.g. autonomous shopfloor planning may increase cost-effectiveness in one company, but not in another with different framework conditions). This also makes it difficult to derive generalised maturity levels for a shopfloor, as a higher CPPS or technical application level is not generally better or more economical.

Conclusion & Gaps:

Although the paper by Romero-Silva and Hernández-López [69] was a simulative and experimental approach focusing on the specific use case of production scheduling, this finding can certainly be used to conclude that the benefits of implementing a partial or full CPPS depend on the characteristics of the shopfloor. Not every use case described in Section 3.1 provides practical added value for every type of shopfloor. The same applies to the technical framework described in Section 3.2. Thus, a classification of shopfloors as carried out by Romero-Silva and Hernández-López [69] seems to be necessary in order to assess their need for data drivenness/datafication. As a result, it can be hypothesised that the maximum level of maturity described in Section 3.3 does not seem to make economic sense for every company in terms of infrastructure and application. With regard to very specific approaches such as those of Lee et al. [35], it should be noted that, on the one hand, they relate to a specific application (here, process and design planning) and are therefore not holistic and, on the other hand, although they are based on cost–benefit evaluations, they already require a DTS in order to be able to carry out the necessary simulations.

4.2 Organisational framework and implementation

In addition to the pure cost–benefit assessment, other organisational conditions play a role in the implementation of a CPPS/DTS. As these are complex models and systems, the expertise of different departments must be taken into account as a basic requirement. This means that the necessary knowledge must also be available within the organisation and must be pooled in order to identify suitable application scenarios and derive the associated functional components and development tools. This is the only way to build a complex DTS through information fusion, multi-scenario interaction and multi-scale association [37, 76, 90, 91]. For example, in addition to pure domain knowledge, Dittmann et al. [92] mention knowledge of OT and IT as a requirement for implementing DTs. As data acquisition in particular is a challenge in the implementation of DTs, a generic four-step methodology combined with IT/OT knowledge is described to successfully implement DTs in terms of data acquisition. In terms of an organisational approach, participative leadership, continuous improvement process (CIP), organisation and structure, agility, attractiveness, and a highly flexible production flow are the functional mechanisms of the future shopfloor organisational design according to Bader et al. [93] combined with six derived criteria for an agile shopfloor. This is also what the future organisation and employees want. An interesting finding by Warnhoff and de Paiva Lareiro [94] in this context is that ‘low-skilled workers’ (e.g. assembly workers) often have a high motivation for work-integrated (= informal) learning, but that highly structured work regimes and hierarchies tend to restrict this. Contrary to the common opinion of supervisors, this group of workers wants to participate in the digital transformation, but there is often a lack of opportunities for skills development. Operator involvement and empowerment also play an important role in the implementation of data-based systems to ensure that they are accepted and used effectively [95,96,97,98]. Similar to the technical maturity models discussed in Section 3.3, there are also organisational and human-centred approaches. Kandler et al. [89, 99] have developed such implementation and acceptance models for dSFM and the underlying elements that should contribute to dSFM implementation.

Conclusion & Gaps:

For example, Jia et al. [37] confirm that specialised knowledge is required to implement a DTS and, hence, a holistic CPPS in an organisation. General prerequisites such as a supportive learning environment (e.g. hierarchical coaching, training, learning-to-learn ability) and the empowerment of shopfloor workers (e.g. to become experienced data users, data managers, systematic problem solvers and decision makers) have already been discussed in the literature and established on the basis of case studies at a high level of fluidity [98]. However, it has already been confirmed in Section 3.2.2 as a research gap that there is still a need to map a holistic view of various multi-scenario use cases in a complex system as part of an SDT. On this basis, it is also necessary to ask what the organisational requirements for implementation are. This aspect has not yet been investigated in depth and should therefore be considered as a research gap. Although there are existing approaches with a generic character, such as the one mentioned by Dittmann et al. [92], they discuss a specific problem (here, data acquisition) at a very granular level and only present some organisational challenges in a use case-specific manner. The same applies to the general organisational embedding of data-based approaches in order to overcome the problems described by Warnhoff and de Paiva Lareiro [94] that organisational and hierarchical frameworks prevent participation in skills development. Ensuring social connectedness (including data sharing) and breaking down data silos must also be considered as organisations become more process oriented (i.e. value stream orientation) [100]. The lack of process/value stream orientation is a performance gap in many approaches. Furthermore, maturity and acceptance models that focus on social factors, such as those developed by Kandler et al. [89, 99], particularly neglect technical factors, specific use cases and their conceptual implementation and design.

To effectively benefit from Industry 4.0, CPPS, DTS or dSFM, people are of central importance. If the human factor is ignored during implementation, technological projects are doomed to failure, as people are not replaced, but rather transformed in their roles, tasks and socio-technical environment. In this context, we speak of the Operator 4.0, the Digitally Enhanced Operator (DEO) or the Augmented Operator [101].

5.1 Technology-people interaction

As mentioned in Section 3.1, the use cases mentioned therein can be summarised under the keyword ‘decision-making’. Listl et al. [54] see decision-making applications as part of a decision support system (DSS), which is visualised in Fig. 9. This helps people to solve complex problems on the basis of a DT in combination with services such as AI. However, the needs of different stakeholders must also be taken into account; for example, short-term machine failures require different data granularity, information channels and proposed solutions than production planners for large and complex production units. This means that the application and interaction must be user-specific in terms of task, knowledge level and authorisation [102]. Jia et al. [37] confirm this aspect in relation to the fact that even within a single DT, there are different user and target groups for the value of a DTS than for the identification of the appropriate user or target group for the respective DT element (e.g. process, product, production), as they have different interests, e.g. with regard to analysis levels (micro, macro, meso), scenarios or applications (see Section 3.1). Regardless, digital solutions play an essential and, sooner or later, unavoidable role in a DSS due to the increasing complexity and demand for real-time data and decisions [26].

Decision support framework for the shopfloor [54]

Full size image

In addition to the need to adapt a DSS specifically to the operator, it is also necessary to look at things from the opposite angle. Particularly in a human–machine collaborative shopfloor, the positive flexibility of an operator can have a negative impact on the DSS. Specifically, unpredictable human activities can have an impact on the DSS and consequently lead to undesirable (possibly even autonomous) decisions [53]. To address this, Shi et al. [74] have added a social layer to the CPPS where autonomous human–machine collaboration takes place, linked to the cyber layer via knowledge support and feedback. Knowledge management systems (see Section 3.1.6) in combination with assistance systems are useful tools [60].

If we look at individual use cases from Section 3.1, how the decision-making applications interact with the operators is also partially described. For example, the condition-based maintenance approach of Mourtzis and Vlachou [39] or the live monitoring approach of Prathima et al. [27] describe what the operator has to do, how his monitoring system is organised and which tasks change or become more efficient as a result.

Conclusion & Gaps:

Although it is recognised that data-based decisions can be adapted to different stakeholders, for example, or that hybrid approaches are conceivable, this is also confirmed as a research gap [54]. The same applies to the combination of a DTS with a DSS and AI, where the interaction between the worker and the DSS plays an essential role in order to avoid inconsistencies within the DTS and the associated incorrect decisions [53]. In certain use cases, interactions with operators are discussed, but only in the context of that use case. However, it was found that digital visualisation is required to take advantage of (real-time) data for decision-making [11, 25]. The idea of Shi et al. [74] to implement a social layer in the CPPS is mainly discussed from a technical point of view. The human factor and its holistic interaction on the shopfloor with data (e.g. regarding analyses) and data-based decision suggestions is very under-represented in the literature. This applies not only to stakeholder management, but also to the resulting changes in operator tasks and required skills, for example.

5.2 Role change

Building on Section 5.1 on technology-people interaction, the issue of role change needs to be considered separately because of its importance to this topic. The changing interaction between technology and humans in the context of a DSS on the shopfloor is also inherently linked to changes in the roles of operational managers and shopfloor workers. For example, Waschull et al. [91] examined the MES-human interaction. This includes the creation and analysis of DTs, data collection, participation in data-driven control loops and decision-making processes, and other activities around an evolving DT. As a result, the role of the shopfloor worker will move into the domain of the DT (e.g. DT creation and maintenance), partially replacing traditional physical shopfloor activities, which, in turn, will lead to increased complexity and skill requirements, as discussed above.

Warnhoff and de Paiva Lareiro [94] and Szalavetz [103] have also made interesting findings through studies/experiments. The findings were that, on the one hand, data-based/digital assistance systems lead to greater systematic autonomy, but that, on the other hand, they restrict the autonomy of the worker and thus informal learning, which primarily affects low-skilled workers (e.g. assembly workers). These workers may become dependent on static assistance systems integrated into the workspace (e.g. pick by light or step by step instructions), which leads to a devaluation/deskilling of the job/role. On the other hand, ‘highly skilled workers’ (e.g. shopfloor supervisors) benefit from such systems because they can use the system as a decision support tool, thus supporting informal learning. This can lead to a ‘digital divide’ within the workforce if low-skilled workers become increasingly reliant on data/digital systems and skills development in the form of informal learning suffers while high-skilled workers benefit. However, managers can counteract this by ensuring that varied and high-quality activities are carried out or added in parallel when introducing data-based systems, which may require a redesign of the role/organisation. Empowering workers both to access, interpret and use data-based systems effectively and to become ‘decision makers’ to solve problems systematically and independently based on data can counteract deskilling. However, this requires, on the one hand, reskilling and upskilling (e.g. through coaching and IT training) and, on the other hand, the provision of worker-centred support systems (e.g. tailoring data and information flows and IT systems to the worker rather than the manager) [11, 90, 95, 96, 98, 102, 104, 105].

Conclusion & Gaps:

In general, there is a strong focus on technology and less on the human factor in the literature [90]. The majority of articles focus on a very specific topic such as MES-human interaction by Waschull et al. [91], a human-centred KPI approach by Hellebrandt et al. [95] or a worker-centric empowerment approach to digital design by Leyer et al. [96]. Although the changing role of the worker is described in the literature, e.g. in terms of required competencies and changing tasks, there is no holistic approach to tasks, authorities and responsibilities (TAR) in relation to a CPPS/DTS and technical use cases. A model designed by Pinzone et al. [101] for assessing the maturity level of the DEO based on three areas of analysis (shopfloor characteristics and work organisation, decision-making and situation awareness, technology support) provides evaluative approaches, but no solution proposal for the conceptual design of the shopfloor. In addition to the maturity level itself, a target concept for the socio-technical shopfloor development (e.g. Where/how is domain knowledge used? Maturity level of worker roles and worker-centred systems? Degree of autonomy: fully autonomous vs. decision support?), the future role of low-skilled and high-skilled workers (e.g. TAR) and the technology-people interaction must also be available. This includes aspects such as those elaborated by Warnhoff and de Paiva Lareiro [94] and Szalavetz [103] on the digital divide or the requirements for worker-centred empowerment (information, resources, support and opportunities) presented by Leyer et al. [96] including the interdependencies between the two. The organisational embedding of data-oriented systems needs to be considered in direct relation to this, as described in Section 4.2. This, in turn, must be adapted in such a way that the data recipient on the shopfloor also has the necessary opportunities to acquire the competences for the role.

Based on the conclusions and research gaps identified for each theme in the previous chapters, the research questions posed at the outset are now answered and the potential for future work is discussed.

RQ1: What is a data-oriented SFM?

It can be summarised that data is the basis for generating information, which, in turn, contributes to decisions, and decisions are fed back to the shopfloor as part of a feedback loop [65]. Based on this simplified logic, the purpose and use process of data can be summarised in Fig. 10. Whether the information generated from data leads to autonomous decisions, human decision support or hybrid decisions is irrelevant when generalising the purpose of shopfloor data, as all variants can generate added value in individual cases. The same applies to the recipient of the shopfloor feedback. This can be the worker, the machine operator or the machine itself, which can feed the feedback into the shopfloor system. Manual, hybrid or autonomous approaches are possible for both the input feedback and any downstream testing, again representing different levels of maturity.

Purpose of shopfloor data to create a feedback control loop

Full size image

This also explains why the use cases already researched and mentioned in Section 3.1 either promote decision-making as a concrete goal or at least contribute directly or indirectly to decision support, as this can be seen as the main purpose of data use on the shopfloor.

Returning to the question of defining a data-oriented SFM, it should be noted that the maturity models analysed in Section 3.3 and the literature beyond do not provide a clear, consistent and uniform definition of what is meant by a dSFM or a data-oriented SFM or how they are clearly differentiated. Rather, it is a qualitative categorisation based on the degree of maturity that can be achieved. However, with respect to the distinction between a dSFM and a data-oriented SFM, it can be concluded that the literature analysed in the context of application and infrastructure development also addresses issues that are not directly related to the original and subjective understanding of data orientation. Examples include mobile apps, voice input or smart devices, which can be used as input tools to replace manual written input in certain circumstances, to collect data more cost-efficiently or to make something (e.g. a visualisation or meeting) more vivid and interactive [9, 65, 87, 88]. In Section 3.2.1, there are also different types of DSs. In these examples, the ‘what’ may not differ at all in terms of data content, only the ‘how’ in terms of type or medium of input/output (e.g. based on ISO 9241). This distinction is particularly important because a digital ‘how’ is not universally better. For example, paper and other analogue media may still have advantages on the shopfloor, such as the ‘power of the pen’ syndrome (e.g. for problem solving or improvement initiatives). Nonetheless, analogue media have limitations when it comes to using real-time data, for example in decision-making, so there is an undeniable overlap between data and digital solutions [11, 25, 26, 106].

From this, and from the purpose of shopfloor data illustrated in Fig. 10, it can be concluded that the commonly propagated ‘next generation shopfloor’, ‘digital shopfloor’, ‘smart shopfloor’ or other synonyms (≈ what + how) must be distinguished from a dedicated data-oriented shopfloor (≈ what), despite the lack of a standardised definition. The bottom line, taking into account the results of the literature review, is that the data-oriented shopfloor is seen in this paper as an intersection of both dSFM and traditional SFM. Figure 11 shows the derived conceptual categorisation of data-oriented SFM compared to the traditional shopfloor, where data also plays a role in the status quo, and compared to the digital shopfloor, which is used as a synonym for the future or next generation shopfloor and also includes a broader scope (e.g. apps, smart devices).

Conceptual categorisation of the data-oriented SFM

Full size image

The Venn diagram can therefore also be seen as an illustrative answer to RQ1. As described in the main chapters, data orientation in SFM can be achieved with both digital and traditional solutions. The technical applications described in Section 3.1 fall within the scope of data-oriented SFM. However, a distinction must be made as to whether these applications intersect with traditional SFM or dSFM. The intersection with traditional SFM exists in particular if the applications cannot be mapped digitally for cost–benefit reasons (see Section 4.1) and therefore require manual/analogue data collection. On the other hand, the interface between data-oriented SFM and dSFM exists as soon as digital data acquisition and processing, and thus a higher CPPS maturity level (Section 3.2), is present. In contrast to the technical applications, which may overlap with either traditional SFM or dSFM in addition to data-oriented SFM, methodologies based on the ‘power of the pen’ syndrome described above (e.g. for improvement initiatives) can be clearly assigned to data-oriented SFM and traditional SFM. This is due to the fact that data is used as the basis for the application, but is developed in an analogue and, therefore, traditional way. However, for those applications that need to be captured and used in real-time (e.g. for real-time decisions), a clear allocation to data-oriented SFM and dSFM is possible.

It can therefore be summarised that data-oriented SFM covers all those activities that, as shown in Fig. 10, deal with data generation based on shopfloor operations, data-information transformation and data-driven decision-making, including feedback to the shopfloor. All related areas are within the scope of the data-oriented SFM.

SFM areas that work exclusively with data-independent and traditional approaches (e.g. culture, organisation, on-site management, working principles) or exclusively digitise/visualise input or output in a data-independent manner (e.g. apps, smart devices) are assigned to traditional SFM or dSFM without intersection with data-oriented SFM. These areas are therefore outside the scope of the data-oriented SFM.

RQ2: What is the significance of data in SFM in the manufacturing context?

Data-oriented decision-making on the shopfloor and the associated accessibility and applicability of data and information is the basis for minimising or eliminating superfluous work on the shopfloor [107, 108]. The decision-making applications described as examples in Section 3.1 offer clear benefits that are undisputed in the literature. As already mentioned, the main benefits, depending on the use case, are primarily increased efficiency (e.g. through faster or automated execution of tasks/decisions), reduced costs (e.g. in maintenance), increased effectiveness (e.g. through better decisions) or a general increase in output or performance on the shopfloor. For certain approaches, such as real-time decision-making, digital solutions are essential or bring fundamental advantages (e.g. in terms of efficiency, transparency or timeliness). However, it cannot be generally concluded from the literature analysed that data orientation in SFM does not always and exclusively require digital approaches and has an economic cost–benefit ratio.

RQ3: How should data-oriented SFM be structured?

After deriving the definition of data-oriented SFM, it can be stated that all maturity models that are intended to describe the structure in general are specialised on singular levels of consideration (technology, people or organisation). As a result, there is no clear answer to this research question in the literature, only observations/maturity models on individual sub-areas or solutions for specific use cases/applications. The approaches described in Section 3.3 focus only on the technical area, and there are also separate considerations for the human factor (e.g. according to Pinzone et al. [101]), which peripherally include organisational aspects. A fully comprehensive combination of all three levels of consideration at a higher level of detail (e.g. for different groups of people/stakeholders on the shopfloor and their possibly different levels of maturity/competences/roles), including their interdependencies, is not available in the literature. For example, future skill profiles are described in general terms, but are not discussed holistically with a technical environment and the TARs derived from it. The same applies to possible target images in relation to the three levels, which are discussed separately (e.g. extension of decision support to autonomous decision-making including impacts on the people and organisation). Meissner et al. [87] underline this fact with a study in which about 200 production companies were questioned about their dSFM condition on the basis of the human-technology-organisation levels of consideration. In their summary and outlook, they emphasise the need for a holistic and more targeted development of dSFM approaches, taking into account all three levels, especially for small- and medium-sized enterprises, as larger companies are already more digitised. The aim is not only to present target states and maturity levels, but also to show how the development path can be designed. Approaches such as learning factories are a way to reduce risks such as alienation or incorrect use of systems, test individual prototypes and obtain user feedback on the specific design of use cases [97]. As data-oriented SFM is an intersection of dSFM according to the previously derived definition, this conclusion can also be applied to data-oriented SFM. It should be noted that at certain levels of maturity, digital approaches from the dSFM toolbox are unavoidable (e.g. real-time decision-making) and therefore there is an overlap between data and dSFM. However, in certain cases or at certain levels of maturity, traditional approaches to data-oriented management may be sufficient.

In terms of structuring and conceptualising a target state for data-oriented SFM, the ‘technology’ level of consideration discussed in Section 3 provides the most comprehensive basis from a literature point of view. Although there are still gaps in research in certain technological areas, this level of consideration has been the most thoroughly researched from a conceptual perspective, less so, however, in the context of the other levels of consideration ‘organisation’ and ‘people’ for a possible holistic conceptual design in a data-oriented SFM.

Although the value stream/process-oriented view and design mentioned in Section 3.1 is also becoming increasingly important, the current literature is dominated by the singular consideration of individual data-based systems, metrics or applications. The value stream view of SFM in combination with data and the related maturity levels (technology, people, organisation) therefore represents an area of research with great potential for future work.