Human-centred design in industry 4.0: case study review and opportunities for future research

After excluding the duplicate papers, there are 215 remaining papers whose yearly publication data allow for the extrapolation of two interesting stages from 1997 to the middle of 2020, as portrayed by Fig. 2. First of all, one notices that the topic has gained momentum and research interest in different aspects of HCD. Secondly, for the period 2015–2019, there has been an almost consistent and healthy growth in the number of publications. Obviously, the 2020 data is still incomplete, which shows a lower number of publications than that of the previous years, because this research was carried out in the middle of the current year. Besides, we applied the Hot’s trend prediction method to exponentially conjecture that the research publications could reach 108 papers by the end of 2020. However, the growth rate could be affected due to the global issue of Covid-19.

By examining only 77 included papers, Table 3 presents the most cited papers, accounting for 63% (329 out of 501 total citations). Interestingly, these top-cited papers have almost been published in recent years. This fact shows that the development of HCD has not matured yet, while the scholars have made the references to the recently published papers for new findings instead of citing the previous ones that have not been well generalized in the research community.

The top cited paper of Zheng et al. (2018) outlines future perspectives of smart manufacturing systems in which user experience is considered as one of development challenges, and transdisciplinary research is called for future research. Beyond the technical perspectives, the scholars also drew attention to social aspects. Specifically, the work of Mazali (2018) explicitly concluded that one of the key issues for the future is to design a balance between the worker being able to control the process by using their own intelligence and the automation of digital algorithms. This perspective is also agreed upon by the work of Streitz (2019), who graded the equal importance among humans and technologies in ambient intelligence to achieve the smart paradigm.

By taking a detailed look at 77 included papers, Fig. 3 shows that the most influential countries are accounted for by Germany (18%), followed by Italy (14%), and China (12%). In the regions, European countries have shown strong contributions in the field with 65% publications, which was reflected by several pieces of research—Factories of Future (European Commission, 2013) and Platforms for CPSs (Thompson et al., 2018)—whose recommendation for future research indicates that it has been a long road reaching the systems of HioTL at the matured level together with other emerging technologies. Some specific research programs and priorities in the next three decades are extracted as below:

By examining the journals by which the included papers were published, the transdisciplinary approach of HCD is strongly evidenced by the fact that there are no journals significantly overwhelming other journals. Table 4 reveals two interesting facts. First, the top 11 journals out of 54 journals—which publish 77 included papers—range from varying research disciplines: engineering; computer science; business management; social and philosophy, which is specialized by the journals Cognition, Technology & Work and AI & SOCIETY. This transdisciplinarity integrates cross-disciplinary perspectives—philosophy, engineering, computer, business, and social sciences—in the context of HCD and transcends their traditional boundaries. This fact addresses the interest in extending the research boundaries of various dimensions of HCD: human diversity, physical to cognitive ergonomics, economics, manufacturability, and social and human-related sustainability.

This transdisciplinary approach has also brought different studies across various research contexts, as can be seen in Fig. 4. There are 42 papers out of 77 included papers that clearly indicate their research focuses on specific manufacturing processes and industries: machinery and equipment as the top one, followed by automotive industry and machining process. The adaption of HCD has progressed in more specific fields: adhesive solutions was considered as the case study on which Lee and Abuali (2011) tested their methodology of innovative and advanced PSS; smart labelling design was developed from the foundation of Industry 4.0 human-centred smart label applications proposed by Fernandez-Carames and Fraga-Lamas (2018); design for textiles was implanted with interactive technologies to experiment and enhance fashion emotional design by Wang et al. (2018).

On the other hand, there are 13 papers out of 77 included papers that explicitly adapt HCD in services, for example public service for smart housing services—which seamlessly connect humans and machines—by design for HMI with the application of Bluetooth ubiquitous networks (Diegel et al., 2004) or a 3D-based meta-user interface (Mostafazadeh Davani et al., 2018). For the healthcare sector, Haber and Fargnoli (2019) emphasized the understanding of human needs and proposed the approach of PSS—the integration of products (hemodialysis devices) and services (e.g., technical support, response time)—for the offering’s value. In the same sector, Gervasi et al. (2020) proposed an evaluation framework—which expresses the perspectives of engineering, cognitive, and social science—of HRC to assess the support of robots for elderly people to reach a specific place.

The research methodology is also diverse in both conceptual and empirical research, as evidenced by Table 5. Fifty-six out of 77 included papers (around 73%) take an empirical approach, while the remaining 21 papers (around 27%) contribute to the conceptual findings. Empirical research uses scientific data or case studies for explorative, descriptive, explanatory, or measurable findings, while conceptual research focuses on abstract ideas, concepts, and theories built on literature reviews (Marczyk et al., 2005; Williams, 2011). Those conceptual papers are further categorized into SLR, accounting for four papers (around 5%) that differentiate from traditional narrative review papers (around 22%). The strong point of SLR is a replicable, scientific, and transparent process minimizing bias through exhaustive literature searches of studies and simultaneously providing the traceability of results (Boell & Cecez-Kecmanovic, 2015; Tranfield et al., 2003). Of the 56 empirical articles, 37 papers (around 66%) are qualitative studies and 19 articles (around 34%) are quantitative studies. Those figures explain the current research effort that focuses on describing, explaining, and interpreting HCD is overtaking the research effort on quantification and statistical treatment for supporting or refuting research findings. This fact is reflected by the nature of the social phenomenon being investigated from the human point of view, leading to the difficulty in the generalization of results (Mennell, 1990; Walsh et al., 2015).

Table 5 also reveals the multidimensional approach of levels of research analysis that range from the level of the product to the levels of the workstation, the company and, finally, society. The research on the level of society and the workstation is still modest in comparison with that of the company or the product, accounting for 12 papers out of 77 included papers (around 16%). The figures show there is reasonable space for further research that deals with HCD at cross-layer levels other than the company and product level, which is also consistent with the future research agenda proposed by the European Commission (2013).

In a broader sense, by applying the qualitative research methodology, Fosch-Villaronga et al. (2020) took a step beyond the company level to gather expert opinions addressing social challenges—ethical and legal issues, job availability—due to the use of social robots. They investigated the challenges from both user perspectives—privacy, autonomy, the dehumanization of interactions—and worker perspectives, such as the possible replacement of jobs by robots. Based on the companies’ perspectives with regard to addressing this level of social concerns with the qualitative approach, Mazali (2018) conducted 40 in-depth interviews with managers of 20 manufacturing companies to accommodate the social needs and organizational contexts that involve multiple stakeholders and new roles of intelligent systems in workflows. In the lower area, the company level is addressed by the business cases and processes. For instance, the work of Hammer et al. (2018) shows an extension of existing business models for quality of experience that incorporate user needs and motivation as aspects of the individual dimension. Subsequently, the workstation level concerns the design for human-oriented workstations, for instance, addressed by Gualtieri et al. (2020) who concluded the need to perform an accurate ergonomic assessment at the first phase of workstation design. The last layer of analysis is the product level, whose design object is an artefact or a service solution.

In addition to the transdisciplinary approach—an integration of cross-disciplinary perspectives—in HCD, this multidimensional approach is also evidenced by the cross-layer level—the product and/or service, workstation, company to social level—in which humans are centred.

The strategy to categorize the case studies follows the design concepts embraced by the corresponding paper. Those concepts are not always explicitly indicated by the papers that may use the term “human” or “user” and even consider them interchangeable terms. This confusion is also reported by Holeman and Kane (2020) and Bazzano et al. (2017). Therefore, Table 6 structures the description of the concepts associated with their common context of use.

The variants of HCD reinforce the findings of the transdisciplinary and multidisciplinary approach—physical to cognitive ergonomics, products and/or services to social-technical systems—towards human diversity, ergonomics, economics, manufacturability, and social and human-related sustainability. Based on the understanding, Table 7 captures the emerging trend that provides insights into six concepts summarized in chronological order.

The top three concepts—namely HCD, PSS and UCD—that account for 35 out of 43 case studies (around 81%) are the most frequently and recently used concepts during the last three years. HCD is the most popular term, although it originated somewhere in the 1400s to systematically improve design for procedures and tools to accomplish the work (Nemeth, 2004). HCD has changed dramatically in the context of Industry 4.0, where scholars have expanded the research of physical ergonomics to systems including humans. Specifically, the case studies are designed in various implementation scales in different contexts: the product level by testing the method of individual product innovation design in solving bicycle problems based on ergonomic perspectives (Wu et al., 2013); the company level by validating the proposed model of the artificial self-organizing manufacturing control system explicitly putting humans in the centre of the system design (Pacaux-Lemoine et al., 2017). Beyond technology, the trend of market personalization has received increasing attention from researchers. The literature witnesses the increasing number of case studies that pertain to the concepts of PSS and UCD. The case studies also distinguish clearly between PSS and UCD by the way that PSS focus on business models at the company level while UCD experiments focus on human experiences about design for product and/or service solutions at the product level in consideration of human diversity and social aspects.

On the other hand, the case studies related to the concepts of HioTL, HMI and HRC are not well accounted for. One of the technical challenges is that the boundaries between technologies and humans are increasingly fuzzy: language processing, social robotics, artificial intelligence, cyber physical systems, virtual reality, and augmented reality. This phenomenon is blurring the limits of where the human ends and technology starts (Frauenberger, 2019; Gervasi et al., 2020; Weichhart et al., 2019; Wojtynek et al., 2019). Moreover, recent research tends to focus on technical aspects instead of tackling existing problems related to error-prone interaction between human and non-human actors (Klumpp et al., 2019; Song et al., 2016).

Another fact shows that the research community has responded in a determined way—35 case studies during the period of 2017–2020, which greatly exceeds other periods—to the call for empirical research in the field (Benabdellah et al., 2019; Kadir et al., 2019). This effort, which is worthy of emphasis, reveals an increasing interest in empirical studies, which brings research and industrial applications closer together. This trend also aligns with the future research recommendations: Factories of Future (European Commission, 2013) and Platforms for CPSs (Thompson et al., 2018). The following deep analysis manifests the design methods connected with supporting technologies that the papers embrace in order to realize the effort in question.

Norman (2016) explains that “the human mind is exquisitely tailored to make sense of the world” (p. 2). This ability requires products and/or services that are designed for easy interpretation and understanding. Therefore, methods for design must define procedures, techniques, aids, or tools to discover the minds of humans—users, customers, stakeholders—that serve as key inputs resulting in well-designed solutions. Figure 5 captures the frequency of design methods that are discussed in four generic groups: discovery, clean-up, engineering, and experiment.

Around 63% of case studies make the most of iterative design: knowledge obtained through the discovery is assured by an iterative process of idea exploration, gathering, and assessment. This method contains a bundle of procedures, techniques, and tools—participatory design, interviews, questionnaires, focus groups, scenario observation, field studies, prototyping—for searching and matching design ideas with the human mind. These approaches help designers focus on human diversity to gain critical design inputs and feedback: requirements elicitation acquired from maintenance professionals by field studies (Kaasinen et al., 2018), human perception of different stakeholders by focus groups (Turetken et al., 2019) and usage difficulties of non-expert users by scenario observation (Song et al., 2016). On the basis of questionnaires, Kong et al. (2019) also studied and called user frustration “the key pain spot” in the context of industrial wearable systems. They also pointed out countermeasures—confinable and reconfigurable modularized hardware sets—that address the usage, cognitional, and operational issues, and reduce the complexity and cost in the design solutions considering various aspects: ergonomics, plug-and-play features, and manufacturability. The modular approach is also comparable to product platform design that tackles the issues regarding manufacturability—product customization, variety, and commonality between products—and brings a competitive advantage: reduction in design effort and time-to-market for future generations of products (Farrell & Simpson, 2003; Martin & Ishii, 2002; Simpson, 2004). This is further evidence to show the necessity of the transdisciplinary and multidimensional approach within which an engineering method can also be applicable in the context of HCD to integrate human and non-human factors: human diversity, ergonomics, economics, manufacturability, and sustainability.

In addition to the acquisition of human needs and requirements, iterative design is also suitable for investigating “what-if” scenarios on design solutions. For instance, Kymäläinen et al. (2017) and Harwood et al. (2019) built fiction prototyping—video-illustrated and tangible interaction tools—to facilitate human-centred perception and cognition of the future potentials of products and/or services. This so-called design fiction—an interactive and tangible approach—evaluates alternative design solutions or criticizes existing ones (Knutz et al., 2014) before they are manufactured and/or delivered to customers, which enhances the robustness of iterative design by deeply understanding human experience.

Even though an effective understanding of human requirements is vital for well-designed solutions, this task is difficult due to various subjective human ideas: prioritization, complexity, imprecision, and vagueness. Clean-up is significantly more challenging for requirements of services than those of products (Haber & Fargnoli, 2019; Song & Sakao, 2016). To respond to the challenge, 6 out of 43 case studies (14%) deal with fuzzy inputs and multiple-criteria decision making by applying mathematical models: analytic network process (ANP), Thurstone’s Law of Comparative Judgments (LCJ), fuzzy set theory, and geometric vectors. While Zhu et al. (2015) took advantage of ANP to determine and prioritize the importance weights of engineering characteristics derived from a set of different customer requirements, Haber and Fargnoli (2019) prioritized customer requirements by the LCJ that transforms the customer preferences into scale values and then represents the importance of each preference. To quantify the complexity, Mourtzis et al. (2018) proposed a 2D geometric vector to estimate the product and service’s design complexity, which is defined by information content, quantification of information, and diversity of information. This quantification of complexity supports the decision-making process on alternative design solutions, taking manufacturability into account. To deal with imprecision and vagueness, Chen et al. (2016) evaluated the users’ perceptual images and feelings about products by the use of the fuzzy membership degree of emotional semantic descriptive words (e.g. traditional-modern, geometrical-organic, romantic-realistic). They also used a statistical method—principal component analysis—to cluster the varying user perceptions and feelings into homogeneous groups of design characteristics. Similarly, Leng and Jiang (2017) clustered similar individual service design processes into homogeneous bundles of services by applying a granular computing method—fuzzy set theory combined with quotient space theory for classification (or clustering) of uncertain complex problem (Zhang & Zhang, 2010). Taking both customer and engineering subjective ideas, Chen (2016) carried out the fuzzy analytic hierarchy process (AHP) to develop good quality design based on the imprecise relationship between engineering experience (robust design, design optimization, design cognition) and customer experience (requirements management, ergonomics design). Based on that, the author also proposed a linear programming model to optimize the total profit of the product mix-experience portfolio, taking economic considerations into account. This cost–benefit analysis needs to be embraced because its importance is stated by several authors, especially with regard to the entire life-cycle cost analysis (Anke, 2019; Heidari et al., 2020; Rodriguez et al., 2020). These mathematical methods are useful in dealing with the multiple-criteria decision making and fuzziness (uncertainty) under their own assumptions, constraints, and computing capability, requiring practitioners to be transdisciplinary and understand properly the methods in their context of use. For references regarding these methods, refer to the work of Golden et al. (1989), Kubler et al. (2016), and Liu et al. (2020).

In addition to the discover and clean-up, 26% of the case studies apply human factors and ergonomics to understand and evaluate quantitatively the interactions—physical and cognitive ergonomics—among humans and other actors (e.g., design artefacts, virtual objects, system interfaces, industrial workstations) from the engineering perspective. This method is not only for the expected cost saving, but also for the higher process efficiency that can be realized by shedding light on human factors and incorporating human needs and behaviour in a healthy, safe, efficient and enjoyable manner (Labuttis, 2015; Soares & Rebelo, 2016). In the context of Industry 4.0, this method is also supported by the digital technologies—virtual and mixed reality, eye-tracking systems, digital modelling and simulation for virtual workplaces—to facilitate designers to capture and analyse design data that span from the physical to cognitive level. On the cognitive level, Wu et al. (2016) studied the relationship between interface complexity and user diversity—novice and expert (human background)—by measuring users’ psycho-physiological data (eye-movement research) combined with questionnaire evaluation methods: NASA-task load index and Questionnaire for User Interface Satisfaction (QUIS) to measure operators’ subjective feelings and workload throughout the experiment. These eye-movement data provide insights into the visual, cognitive, and attentional aspects of human performance (Duchowski, 2002). In addition to the psycho-physiological analysis, Richert et al. (2018) surveyed participants’ personality dimensions—agreeableness, conscientiousness, neuroticism, openness to experience—to measure the performance and human perception of hybrid human–robot collaboration. On the physical level, Caputo et al. (2019) carried out an appraisal for the human-centred workplace design by reproducing a virtual workplace in which digital human modelling simulates the whole human task towards preventive ergonomics. Peruzzini et al. (2019) also designed the virtual workstation with preventive ergonomics by the use of digital technologies: virtual and mixed reality. They also used questionnaire methods to quantitatively measure postural comfort: Rapid Upper Limb Assessment (RULA) and Ovako Working Posture Analysis System (OWAS). The case studies apply a wide range of assessment methods regarding human factors and ergonomics: from simple checklists to more complex techniques; from physical ergonomics—for human use and performance (e.g., musculoskeletal symptoms, body posture, low back disorders)—to cognitive ergonomics—for human perception and cognition (e.g., mental stress, emotional stress, situation awareness). In addition, the work of Tillman et al. (2016), Forsythe et al. (2017) and Dalle Mura and Dini (2019) provides a good source of numerous methods for human factors and ergonomics that allow for achieving the various objectives of both manufacturability and social sustainability.

To bridge the gap between human requirements and engineering characteristics, four out of the 43 case studies apply quality function deployment (QFD), which originated in the automotive industry and has been being used with different applications in diverse fields for five decades (Kowalska et al., 2018; Zairi & Youssef, 1995). This method identifies human-centred requirements, classifies the importance of those requirements, defines engineering characteristics that may meet those requirements, allows for verification of design conflicts among them, and then prioritizes design solutions. In the analysed case studies, this method is also integrated with different methods—application space map and innovation matrix (Lee & Abuali, 2011), ANP (Zhu et al., 2015); AHP, fuzzy AHP, entropy weight method (Ma et al., 2017); LCJ and Kano model (Haber & Fargnoli, 2019)—to enrich the prioritization and segmentation of the design requirements. The requirements after the cleanup are further converted into the engineering parameters by the QFD. For further reading, the work of Chan and Wu (2002) and Prasad (1998) may be of interest to the reader.

Furthermore, other methods also include the Kano model, Kansei engineering, business process modelling, and notation (BPMN). While Haber and Fargnoli (2019) applied the Kano model to prioritize and classify customer requirements into four different categories—must-be, one-dimensional, attractive, indifferent—for the segmentation of customer value propositions, Wang et al. (2018) parametrically linked the customer's emotional responses—physical and psychological—to the properties and characteristics of a product and/or service. If these methods focus on a particular process in design (requirement elicitation converted into engineering characteristics), Prinz et al. (2019) highlighted the use of BPMN to represent workflows—a graphical modelling language for all kinds of business processes. The BPMN is useful for examining a graphical description of design processes to different levels of granularity and discovering inconsistencies and/or differences in sequential steps, conflicting names, or acronyms, to name a few. Even though the methods have only been mentioned one time by the 43 case studies, they have been adapted and applied by different fields for years. Several publications are interesting works that may help readers have a better idea about the Kano model published by Zhao et al. (2020) and Shahin et al. (2013), Kansei engineering reviewed by Shiizuka and Hashizume (2011) and Coronado et al. (2020), BPMN studied by Ko et al. (2009) and Chinosi and Trombetta (2012).

Lastly, another way of gaining knowledge in design is empirical experiments, which account for four out of the 43 case studies. This method is useful for understanding what-if scenarios by different design configurations: an assisted versus collaborative robotic system that supports workers in a plug-and-produce workstation (Wojtynek et al., 2019), an automatic speed versus adaptive cruise control system for pedagogical learning supports (Vanderhaegen, 2019), delivery of health care services for seniors between a community hospital and social service agency (Hoe, 2019), augmented reality that supports trainers versus trainees in phone repairing operations (van Lopik et al., 2020). Those empirical experiments allow for designing hypotheses and gaining knowledge by means of direct and indirect experience. However, this method requires knowledge of the experimental setup and validation; it also has limited generalization of results due to controlled settings (Kulyk et al., 2007).

In summary, the case studies apply various methods that are categorized in the four generic groups—discovery, clean-up, engineering, experiment—associated with supporting technologies to tackle different problems, which requires the transdisciplinary approach for understanding and applying the methods in their proper context of use. While iterative design is power in discovering the human mind (needs, perception, cognition), mathematical models prioritize and classify those human inputs and support the decision-making process on design alternatives. Furthermore, human factors and ergonomics enrich the understanding of interactions—physical to cognitive ergonomics—among human and non-human actors with the support of digital technologies: virtual and mixed reality, eye-tracking systems, digital modelling and simulation for virtual workplaces. To convert the voice of humans into engineering parameters, the case studies have diverse approaches—QFD, Kano model, Kansei engineering, BPMN—and are used in different combinations. Finally, the empirical experiments gain knowledge based on the investigation of what-if scenarios under the human perspective, which is useful for iteratively improving and testing design solutions. Besides, researchers and practitioners alike also benefit from other relevant engineering methods—product platform design (Simpson et al., 2014), design for manufacturability and concurrent manufacturing (Anderson, 2014), to name a few—that embrace the transdisciplinary and multidimensional approach to deal with a multi-objective design problem towards human diversity, ergonomics, economics, manufacturability, and sustainability.

These various methods dealing with different problems in diverse contexts of use lead to different lessons learnt in the form of their research results and limitations. The following lessons learnt are useful for subsequent researchers to choose proper research areas and advance research contributions to the field by avoiding the research limitations.

One way to organize the case studies sharing mutual facts and document them as the lessons learnt is to use an affinity analysis, which is also known as the KJ method and applied in various fields (Awasthi & Chauhan, 2012). The information captured during the analysis is tabulated by “Appendix” (Table 11), providing researchers useful details about design for case studies. Based on the analysis output, Table 8 categorizes the case studies’ results and limitations into six groups of research results (RR) and four groups of result limitations (RL).

One of the most attractive outcomes those case studies reported is the exploration of the design success factors—which are denoted as RR2 accounting for around 47% of the case studies—revealing how the successful deployment of design oriented to humans can be generalized in various contexts. Figure 6 structures those success factors as a triangular decision-making diagram:

The knowledge management of these design data is well expressed as an enabling success factor that can be exploited by digital technologies. These technologies facilitate the collection, organization, retrieval, and reuse of design knowledge in an effective manner. While Fu et al. (2019) took advantage of IoT solutions (sensors) for user data collection—unintentional behaviour, emotion, culture—and artificial intelligence for data processing, Vanderhaegen (2019) and Grandi et al. (2020) made use of digital and mixed reality simulation in measuring human factors—physical stress, physiological data—and evaluating their design experiments. Instead of starting from scratch, Zhu et al. (2015) and Leng and Jiang (2017) established mathematically a collection of semantic commonalities derived from historical design ontology-based databases—activities, functions, concepts, process sequences—to build a knowledge platform from which a stream of new derivative products and services can be efficiently developed. The objective is to design for variety and custom solutions, enabling designers to not only save time and cost but also make the most of the experience and expertise that were dedicated to the past design activities. The method used to build the knowledge platform is also comparable with product platform design, which has been maturely researched over the last decade (Simpson et al., 2006, 2014) and is a useful source regarding methods and applications for researchers in the field of product and/or service design.

The second group is the engineering objectives of design (RR1) that are converted into key performance indicators to quantify the effectiveness of the proposed models or frameworks. Around 23% of the case studies indicate that their proposed solutions achieve the engineering objectives: avoidance of ergonomic risks (Caputo et al., 2019; Ceccacci et al., 2019), improvement of productivity and simultaneously biomechanical workloads (Gualtieri et al., 2020; Wojtynek et al., 2019), production performance in terms of quality and engineering time (Pacaux-Lemoine et al., 2017; Prinz et al., 2019). Furthermore, Wu et al. (2013) proposed a multi-function and modular method for design focusing on human anthropometrics—the branch of ergonomics that deals with measurements of the physical characteristics of human beings (Pheasant, 1990)—and extending products’ service life towards sustainability. Similarity, Chen et al. (2016) applied a clustering method for product family design based on anthropology—research in understanding human culture, society, and difference (Monaghan & Just, 2000)—to improve the agility of the design process towards manufacturability. This product family design allows designers to not only utilize existing design methods from the product platform to form a series of products, but also gain inspiration from different ethnic groups—human diversity with distinct cultural traits—to extract ideal design elements. In another aspect, Chen (2016) emphasized directly the cost–benefit analysis of design quality, taking into account two economic elements: estimated profit; total cost comprising R&D cost, market capital, and design quality for market share. The reported figures prove the robustness and performance of a system—human diversity, ergonomics, economics, manufacturability, sustainability—can be achievable with the approaches of HCD.

The next research interest is to provide supporting design frameworks (RR6) that facilitate the design process by providing systematic thinking—the use of the integrated novel design methods (innovation matrix, application space mapping, QFD) and Lean initiatives (avoidance of valueless reworks and activities)—towards economic sustainability (Lee & Abuali, 2011; Pezzotta et al., 2018). Other studies focus on design solutions for complexity and uncertainty: incomplete information regarding human requirements (Haber & Fargnoli, 2019); the changes in human preferences (Lin, 2018); decision making on different design alternatives for mass customization towards manufacturability (Mourtzis et al., 2018); interaction requirements among non-human—smart manufacturing devices/tools, core enterprise business systems (ERP, SAP)—and human actors (manufacturers, designers, users) (Mostafazadeh Davani et al., 2018; Song et al., 2016; Zhang et al., 2020); adaptation of design processes to the context of small-and medium-sized enterprises (Adrodegari & Saccani, 2020; van Lopik et al., 2020). These studies tackle different problems scattered across life-cycle design phases, useful to consider in relation to further research to address the relevant problems in a comprehensive way.

Around 12% of the case studies made an effort to validate the effect of human diversity on the design outcomes (RR3). They concluded with the important inclusions of individual differences—background, age, gender, education, cultural influences, privacy management—in design. Statistically, Wu et al. (2016) confirmed that information overload in interface design increased cognitive workload for novice operators compared to expert operators and therefore decreased user efficiency. Similarly, Van Acker et al. (2020) concluded statistically that higher acceptability of wearable mental workload monitoring was associated with being a woman (for trust in the technology), higher technology readiness—the willingness to accept new technologies and security about private data (Victorino et al., 2009)—and lower educational backgrounds. Besides, lack of considerations regarding specific classes of difference between humans leads to major effects on design outcomes in various design contexts: age with older people (aged 55–75 years) in safe driving (Jung et al., 2017) and health sector (Hoe, 2019); cultural influences (Russians, a Frenchman, a Chinese) in the experiment of long-term isolation in a limited room space (Boy, 2018). These studies address the concern that if design does appreciate individual differences towards the multidimensional approach—considering not only product and/or service design but also social aspects—this could avoid the thwarting of all research efforts and the subsequent lessening of potential benefits.

In addition to the multidimensional approach, four studies also directly address the need for collaborative design frameworks (RR5): the transdisciplinary approach during the life-cycle design phases. Ma et al. (2017) exploited common expertise of transdisciplinary teams to convert customer requirements into semantic requirement groups that were subsequently transferred into product design specifications through the use of QFD. Based on the perspective of cross-cutting collaboration for advanced business intelligence, Kong et al. (2019) structured a common platform design of wearable-enabled applications with three aspects of manufacturability: re-configurability, robust architecture, and design scalability. This platform allows standardization by taking advantage of plug-and-play features and modular approaches to integrate human and non-human actors: artificial intelligence, virtual reality, IoT, cloud computing, and cloud-based cyber systems (enterprise resource planning, manufacturing execution systems, warehouse management systems). In addition to manufacturability, Anke (2019) and Turetken et al. (2019) addressed directly the aspects of life-cycle cost analysis in the context of smart services. Specifically, Anke (2019) assessed the profitability of a smart service at an early stage of service design by developing a web-based tool prototype by which project teams from different disciplines collaborate in the design and evaluation process. In a broader sense, Turetken et al. (2019) promoted the transdisciplinary and iterative approach in which a network of actors—providers, customers, authorities, retailers, event organizers—co-creates the value-in-use for customers and generates benefits—financial and non-financial characters—for all network partners moving towards sustainability. Each study focuses on an important aspect of design—human diversity, ergonomics, economics, manufacturability, sustainability—that needs to be considered together in a transdisciplinary and multidimensional approach for future research.

In the last group of research interest, three studies present experience-driven approaches that visualize design scenarios (RR4) regarding future possibilities to exploit human experience. Based on design fiction, both Kymäläinen et al. (2017) and Harwood et al. (2019) demonstrated the usefulness of the video-illustrated prototype in avoiding the difficulty of interpreting abstract verbal descriptions of new design. This method enables designers to interactively envisage a spectrum of “what if” scenarios towards human experience that may then be explored by using the range of other design methods: focus groups, interviews, and questionnaires. Besides, Kaasinen et al. (2018) made the most of the technologies in Industry 4.0—wearable technologies, virtual and augmented reality—to visualize the human experience of future maintenance work: feeling competent, feeling connected to the work community, feeling a sense of success and achievement by performing better in jobs. These studies go beyond technical design towards the multidimensional approach: they go from the technical to the social aspects.

Even though all case studies reported positive outcomes, four groups of result limitations are also acknowledged. The most frequently reported limitation is the lack of statistical power in result validation (RL1)—accounting for 60% of total analysed case studies—and the rest is undefined due to limited information for making the conclusion. The lack of statistical power shows limitations in experimental set-up conditions: low sample sizes, lack of fitting in target participants, lack of sound statistical studies, and other biased experimental aspects (Pacaux-Lemoine et al., 2017; Richert et al., 2018; van Lopik et al., 2020). This limitation is followed by the lack of generalizability (RL2) showing the insufficient evidence of the extent to which findings from one study in one context can be applied and reproduced to other contexts. Specifically, 56% of the case studies are constrained and required to be tested by further quantitative methods to prove the transferability of their observed results to other usage contexts (Adrodegari & Saccani, 2020; Haber & Fargnoli, 2019; Kong et al., 2019; Witschel et al., 2019). The next limitation is categorized as incomplete solutions to implement the proposed models (RL4)—accounting for around 30% of the case studies—claiming the quality of the proposed models will depend on other external factors. These factors include the “manual” processing of the proposed models, resulting in application difficulties (Ceccacci et al., 2019; Zhang et al., 2020), which requires additional efforts in further development of supplementary methods and applications to achieve model completion in real contexts (Grieger & Ludwig, 2019; Leng & Jiang, 2017; Lin, 2018; Peruzzini et al., 2019). Finally, around 23% of the case studies do not explicitly provide the validation of effectiveness of the proposed solutions (RL3), which emphasizes the need for future research for their validation in various contexts of usage; otherwise, the practical effectiveness of the proposed solutions from the studies is limited (Ceccacci et al., 2019; Haber & Fargnoli, 2019; Witschel et al., 2019).

These limitations are explained through the evaluation methods—which are different from the design methods used as procedures or processes for attaining research findings—applied by the case studies to validate their corresponding research findings. Figure 7, which is visualized from the detailed data of “Appendix” (Table 11), shows the top four evaluation methods accounted by qualitative methods: questionnaires, interviews, scenario observation, and workshops. These methods validate the effectiveness of the corresponding proposed models by capturing and communicating the participants’ feedback via different means, leading to a potential lack of robustness in research and encompassing subjectivity and bias in research conclusions (Jung et al., 2017; Richert et al., 2018; Van Acker et al., 2020), which is followed by insufficient generalizability, as analysed above.

Although there is a small portion of case studies applying quantitative methods—hypothesis testing and mathematical models (around 9%), performance comparison (around 7%), and ergonomic analysis (5%)—the validation of the case studies’ findings is still questionable. Specifically, by applying the hypothesis testing, L. Wu et al. (2016) made an effort to carry out a case study of eye tracking with 38 participants that compared three levels of interface complexity in LED manufacturing systems, resulting in the statistical conclusion of interface complexity and user background affecting the user experience. However, the study failed to prove sufficient statistical power, showing its proper selection of sample size. Moreover, the sampling procedure included only the participants who were all from the same company, leading to biased results and affecting the generalizability of research outcomes. Out of 43 case studies, Ceccacci et al. (2019) and Gualtieri et al. (2020) conducted ergonomic analysis to validate the effectiveness of their workstation design—productivity, human postural comfort—with a sample size of only two participants. This small sample size, due to its lack of generalizability, requires further research to validate the studies’ applicability in a real context with human diversity. This problem was further evidenced by Van Acker et al. (2020) who reported that, statistically speaking, the replication of their case study’s findings found in the first experiment was not successful in the second experiment carried out within the same research context, so leaving the conclusion inconclusive. These limitations lead to a lack of robustness in research findings and reduce applications of these studies in industry and research alike.

In summary, the research efforts contributing to the realisation of human roles in Industry 4.0 span six groups of research results: exploration of design success factors, achievement of engineering objectives, provision of supporting design frameworks, validation of the effect of human diversity on design, provision of transdisciplinary frameworks, and visualization of design scenarios. Each study focuses partially on its own defined aspects, which provides a useful reference for future research that combines the transdisciplinary and multidimensional approach towards human diversity, ergonomics, economics, manufacturability, and sustainability in a comprehensive way. Besides, it is worth realizing the lessons learnt in order to overcome the acknowledged limitations—limited statistical power in result validation, lack of generalizability of research findings, further requirements for the supporting methods, lack of validation of the effectiveness—and enhance the robustness of the research findings. This will inspire research applications to both industry and research. Finally, the following section discusses the results of the in-depth review and ends with future research opportunities.