1 Introduction

An outcome of the industrial revolution and mass production is the standardization of products. This process has many advantages, including the ability to create many product copies from a single design and the ability to leverage scale to reduce overall product cost. Another advantage is the ability to use standardized parts during system operation and sustainment to reduce the cost of maintenance. While the variability between products is controlled during manufacturing, their use often varies, increasing the differences between products. If these differences are not accounted for, the variability decreases the ability to properly estimate product performance and plan maintenance cycles. Furthermore, specific products within a product family can be different, placing further stress on logistics for sustainment of these products. Digital twin technology has been developed in recent years to address the problems which arise from product and environmental differences (Tuegel et al. 2011). The digital twin concept includes constructing a digital representation or model of an individual product to improve the accuracy of maintenance and performance predictions for individual products (Kobryn 2020). Thus, a digital twin has been described as the model of a component, product, or system developed by a collection of engineering, operational, and behavioral data which support executable models, where the models evolve over the lifecycle of the system and support the derivation of solutions which assist the real time optimization of the system or service (Boschert and Rosen 2016).

Recently, this term has been extended to humans using the term “human digital twin.” This term has been applied in diverse fields, including medicine (Chakshu et al. 2019; Corral-Acero et al. 2020; Hirschvogel et al. 2019; Liu et al. 2019; Lutze 2020), sports performance (Barricelli et al. 2020), manufacturing ergonomics (Caputo et al. 2019; Greco et al. 2020; Sharotry et al. 2020), and product design (Demirel et al. 2021)(Constantinescu et al. 2019; Demirel et al. 2021). Although the human digital twin concept may be analogous to digital twins of products, there are distinct differences, including increased underlying variability between humans and the fact that humans often employ products to achieve their goals. Commonalities can be observed in the use of the digital twin concept across the fields where it has been applied. For example, each of these fields discusses constructing and applying a model of humans which is informed by data collected from sensors which provide insight to human behavior and performance in real world settings. However, there are also differences. For example, in medicine or sports performance, the models often focus on improving the human or human performance while in manufacturing and product design, the model’s focus is on improving either the process or the artifacts with which the human interacts. As a result, the components which are modeled within each of these applications differ.

A clear example of a human digital twin from the literature pertains to manual material handling within manufacturing or warehousing applications (Greco et al. 2020; Sharotry et al. 2020). In this example, sensors are deployed with the human in their environment which monitor the human’s kinematic motion as they perform work within the environment. Other information, such as objects being lifted, or forces required to activate items within the environment are also collected. This data populates simulation models which estimate the fatigue in various muscle groups of the operator. These fatigue estimates support assessments and potential changes in work or rest schedules, handling processes, or material handling tools to improve overall worker health, safety, and productivity. Thus, a model of a human attribute is applied within a closed-loop system which fulfills the criteria for a digital twin (Boschert and Rosen 2016). While models of human behavior, performance, and state have been developed for decades, the key concept of human digital twins involves including these models within a tightly coupled, closed-loop system, where this system permits the model to evolve with data from a specific real-world human as they act within an environment and to provide feedback with the goal of enhancing the performance and well-being of the human to which it is coupled.

While the applications for human digital twin technologies differ, advances in one field may inform application of this technology in other fields. In fact, it is common for models which were developed to understand human performance for medical or sports applications to inform other applications, such as product design. Therefore, the current research seeks to leverage a review of existing literature regarding digital twins, and more specifically human digital twins, to derive a common definition of the human digital twin, a taxonomy of the components useful for construction of the human digital twin, and use cases of this technology within product development and sustainment across the system lifecycle. The overall goal of this paper is to support the development of this technology by proposing common terminology and a vision for applying this technology to support product development and deployment, with an emphasis on product development and deployment within the United States Department of Defense acquisition system. It is hoped that socialization of a common definition and potential applications of this technology within the human factors, ergonomics, and human systems integration communities will lead to synergistic efforts which improve the development of this technology.

2 Method

The current research consisted of (1) conducting a review of literature pertaining to human digital twins, (2) developing a structural model of a human digital twin based upon the components of human digital twins from the literature, (3) applying this literature to derive an inclusive definition, and (4) synthesizing and developing use case models based upon potential applications inferred by the literature.

2.1 Literature review

The literature search included a Google Scholar search on the term “human digital twin” which was complemented by a search on “digital twin” and either “human,” “man,” or “woman.” Each search was limited to dates between 2011 and July 2021. The search on human digital twin returned 99 articles while the complementary search returned 15,000 results. Each of the human digital twin and the abstracts of the first 1000 articles on digital twin was then reviewed. As the purpose of this paper was predominantly to understand the structure and application of human digital twins to system design, development, and operation, the abstract review determined if the articles discussed the creation or application of digital twins which included a model of a human as part of the digital twin. A total of 54 unique papers which included a model of a human either as the digital twin or as a component of a digital twin were identified. For example, a model of a human may represent an athlete where performance is affected by nutrition, rest, and prescribed physical activity while a model with a human as a component may include a workstation in a factory where performance is influenced by factors such as nutrition, rest, and activity of the human but also by attributes of the system in which they are performing, such as material handling equipment and procedures.

Tables 1 and 2 show the application space and the publication year for the resulting papers. As shown, most of the papers either pertained to the manufacturing or medical domains. Four additional papers were focused specifically on understanding and modeling human biomechanics, predominantly with a focus on reducing musculoskeletal injuries in the workplace. Four additional papers focused on health and fitness applications, primarily for sports training. The remaining papers appeared to uniquely address five additional application areas, including general product design (Demirel et al. 2021), military application (Li et al. 2020), disaster management (Fan et al. 2021), organization design (Parmar et al. 2020), and identity management (Zibuschka et al. 2020).

Table 1 Application areas for human digital twins based on literature review
Full size table
Table 2 Publication year for papers in literature review
Full size table

As shown in Table 2, practically all the publications appeared within the past 4 years. The earliest paper does not mention the words human digital twin, but discusses the construction of a digital athlete, which is updated based upon sensed information from the application environment (Alderson and Johnson 2016). The earliest papers using the term “Digital Twin” while referring to human models were published in 2018 with one of these papers appearing in the medical literature (Bruynseels et al. 2018) and three appearing within the manufacturing literature (Graessler and Poehler 2018a, 2018b; Kemény et al. 2018).

2.2 Modeling structure and use cases

The structure and use case models of human digital twins were created using the Systems Modeling Language (SysML). This language provides a standardized language and family of diagrams for representing systems during design and development. The SysML diagrams include diagrams for representing system requirements (i.e., requirements diagram), structure (i.e., block definition and internal block diagrams), and behavior (i.e., use case, activity, state machine, and sequence diagrams) (Delligatti 2013; Friedenthal et al. 2014). Diagrams are also provided for structuring information (i.e., package diagrams) and supporting mathematical modeling (i.e., parametric diagram).

SysML was chosen to represent aspects of human digital twins as the digital twin concept has been described as a method to extend Model-Based Systems Engineering (MBSE) from the design to the operation and sustainment phase of a product (Boschert and Rosen 2016) and SysML has been developed to support MBSE. Furthermore, this language is an object-based language which permits the description of a generic system, sometimes through a reference architecture, and the extension of this description to specific instances or instantiations (King et al. 2020) and has been used to model the human’s contribution to system performance (M. Watson et al. 2017a, b; M. E. Watson et al. 2017a, b). Thus, the generic system description provided through the diagrams in this paper can be extended to illustrate the application of human digital twins in specific applications or instances.

Each of the articles pertaining to human digital twins were analyzed to determine the structural components and use cases they discussed. These elements were then used to compose a SysML block definition diagram to illustrate the infrastructure for a human digital twin. Furthermore, additional components were added based upon additional functions which were implied by the discussion of human digital twins. Similarly, use cases were extracted from each of the articles from the literature review. These use cases were interpreted with respect to product design and application within the United States Department of Defense (DoD). Separate use case diagrams were then developed for the application of knowledge gained from human digital twins to product design and operation. While the use cases are specific to the DoD, which acquires products for its own use, many of these use cases can be generalized to other organizations where the organization responsible for product acquisition and the organization employing the product may differ. Overall, these diagrams define the general components and product development-oriented use cases of human digital twins.

3 Structural definition of a human digital twin

As shown in the block definition diagram of Fig. 1, digital twin systems typically include a real-world entity, i.e., the real-world twin, and a digital twin where the digital twin is a representation of one or more attributes of the real-world twin. The real-world and digital twins are linked through an interchange component such that changes in one twin can produce changes in the other (Barricelli et al. 2020; Grieves 2015; Latif and Shao 2020; Mendi et al. 2021; Sharotry et al. 2020; van der Valk et al. 2020).

Fig. 1
figure 1

Block definition diagram showing primary components of a digital twin system

Full size image

This interchange component permits the digital twin to be updated as the real-world entity acts within its environment. The digital twin then executes embedded models which are informed by the data it receives from the real-world twin to simulate one or more aspects (e.g., structure or behavior) of the real-world twin with the goal of determining changes in the real-world twin which might produce an improvement in the real-world twin.

To illustrate these components, it is useful to review examples. In the medical field, Chakshu and colleagues envision a human digital twin system in which the real-world human twin is outfit with portable sensors to characterize hemodynamic pressure as a function of time in the limbs of a patient (Chakshu et al. 2021). As this data is collected, it is passed to a model which uses these values together with other characteristics of the patient to estimate the pressure at other points within the circulatory system. These estimates are used to detect the presence or changes in an abdominal aortic aneurism. This system includes predominantly the human with sensors, a data processing system which collects the data and updates the model, and the model and prediction systems which compose the digital twin. However, the authors constrain the environment of use to clinical settings in which the patient is at rest.

Sharotry and colleagues discuss a digital twin system in a manufacturing environment which consists of a human conducting material handling tasks (Sharotry et al. 2020). The system includes a data collection module which includes motion capture, biometric suites among other sensors, a data analysis and forecasting model, and a database which contains the data which is collected for analysis. In this system, performance and operator fatigue metrics are provided to the individuals in the environment to support improved manufacturing performance and to help reduce human injury. Furthermore, this model can be used to identify material handling steps which induce substantial fatigue, permitting these steps to be evaluated and redesigned. Within this example, the real-world twin includes the manufacturing environment, the human, and the data collection subsystems. The database provides the interchange component, and the data analysis and forecasting model provides the digital representation or the digital twin.

As illustrated by these examples, human analysts apply the digital twin to conduct analysis, develop courses of action, and then apply the courses of action in the real world. The courses of action identify potentially useful changes in the structure or behavior of the real-world twin. In these systems, the analyst or the interchange component then causes changes in the real-world twin, which ideally improves the performance of this entity. Importantly, the two-way exchange between the digital and real-world twins provides the digital twin the ability to sense the real world, creates an understanding of the world, and acts upon it, completing a process analogous to the perceptual loop (Neisser 2014). However, in a digital twin system, this sequence includes a repetitive sequence with the following stages:

  1. 1.

    Sensors on the real-world twin sense the actions and performance of the real-world twin within its real-world environment, as well as relevant state information about the real world.

  2. 2.

    The interchange component conveys the sensed data to the digital twin.

  3. 3.

    An analysis is performed to determine whether the data is consistent with the digital twin’s current model projections and adjusts or builds a model to explain any differences.

  4. 4.

    The model is applied to create projections of future behavior within a virtual environment.

  5. 5.

    The projection of future behavior is compared to a goal state.

  6. 6.

    Based on this analysis, the system determines if a modification to the structure or behavior of the system is likely to move the system towards a goal state, and if so, selects a modification to the physical element or its behavior which moves the real-world twin towards the goal state.

  7. 7.

    The modification is conveyed to the real-world twin and applied to the real-world twin.

This series of steps is repeated, adjusting the digital models in the digital twin and the real-world twin to move the system towards its goal state. Ideally, the coupling of the digital and real-world twins by the interchange component provides the ability to rapidly create robust models of real-world performance, permit these models to be applied to project future real-world performance, and to improve the performance of the real-world system.

To create accurate predictions, the models represent not only the real-world entity but the environment in which the real-world twin is acting. This fact is not evident in much of the literature as the digital twin concept is applied within controlled environments. Thus, these systems were assumed to be closed; that is, they do not interact significantly with their environment. However, systems are often open, readily exchanging information, energy, and matter with their environment (Blanchard and Fabrycky 2006). Thus, it becomes important to model the environment to understand the entity’s interface with the environment. Specifically, the interface between these components require understanding of the spatial configuration, energy, information, and material transfer between the environment and the system (Jain et al. 2010). However, from an architectural standpoint, the environmental model does not necessarily need to be part of the digital twin. Instead, the environmental model may be separated from the digital twin system and simply be associated with this system. This is a significant architectural decision as the human digital twin system will not necessarily require a broad model of the environment but only requires models which reflect the information, energy, and matter that the system exchanges with the environment. Therefore, the complexity of the environmental model may be constrained if it is constructed to capture the necessary elements as opposed to the creation of a more generic and encompassing model of the environment. The real-world twin must also determine the state of the environment and communicate this state to the digital twin.

Reviewing the literature, different components of the digital twin system are illustrated by different authors. Therefore, it is possible to integrate the concepts across these papers to provide an architectural view of a generic digital twin. Figure 2 shows a more detailed block definition diagram based upon an analysis of the digital twin systems in the literature. As shown earlier, this system is comprised of a real-world twin, a digital twin, and an interchange component. However, alluded to before, the real world and a representation referred to as the virtual world are each associated with this system, where these two entities are linked by their state information.

Fig. 2
figure 2

Block definition diagram showing an exploded view of a generic human digital twin system

Full size image

Within the literature, the real-world twin includes a human within the environment. Sensors are used to provide real-time or near real-time information about the human and the environment (Corral-Acero et al. 2020; Sharotry et al. 2020) and this real-time feedback may be augmented with non-real time feedback, such as medical diagnoses, professional electro-cardiograms, or information from imaging devices (Corral-Acero et al. 2020). Additionally, the human can be queried to input subjective information (e.g., mood) or information which is difficult to track (e.g., nutrition information) into tracking applications (Barricelli et al. 2020). In manufacturing and product design scenarios, the human also interacts with physical devices (i.e., system), such as manufacturing equipment, material to be handled, or products which can be instrumented (Barricelli et al. 2020). Additionally, the real-world twin typically includes a process or procedure (Latif and Shao 2020). Although the process is not a physical item, it is important in understanding the behavior of the human. Finally, while not discussed explicitly in the literature, it may also be important for some mechanism to be present to modify the physical environment. The processes may be changed but to change the physical environment, it is necessary to include physical mechanisms, shown as actuators in Fig. 2. As shown, near the head of the arrows associated with these entities, the human digital twin systems, by definition, include a human, one or more sensors, and at least 1 process. The machine and actuators are optional components as they do not exist in all systems, including the sports systems.

Of the three components, the interchange component receives the least attention within most of the published literature. Although the literature is clear that this interchange component permits two-way communication between the real-world entity and the digital entity, its components are not clearly discussed. However, beyond communications, this component can be envisioned as the warehouse for data or the database, although the database is sometimes viewed as part of the digital twin (Latif and Shao 2020). Furthermore, this component can be responsible for analysis to clean the data, fusing information from the various sensors associated with the real-world twin and integration of new data with the existing data (Tao et al. 2019a).

Finally, we can review the components of the digital twin within the digital twin system. The digital twin certainly contains a model of the human. This model can include mechanistic models, which model well understood physical, chemical, biological, and physiological processes, as well as, statistical models which rely upon the data collected from the real-world twin (Corral-Acero et al. 2020). It can optionally contain additional models of the machine or the process which is to be performed (Latif and Shao 2020; Tao et al. 2019a). Also necessary is a prediction engine which generates perturbations to the virtual world, human, machine, or process to understand how changes in these elements affect the performance of the human or the system (Tao et al. 2019a). To assess the outcomes, a performance goal is also required such that the results produced by the prediction engine can be assessed. A mission description is also likely necessary which informs the process model of potential process steps which are considered viable.

Although not shown in Fig. 2, it is also common for a process manager to interact with the digital twin to determine if any of the modifications evaluated by the prediction engine should be applied to the real-world twin. While these modifications may be communicated through the interchange component to the real-world twin, physical changes to the real-world twin may require human intervention.

It is worth noting that during design, the real-world twin may be under development and therefore the real-world twin or even a prototype of the real-world twin may not exist to participate in the digital twin system. Under these circumstances, and perhaps others as well, it can become useful to include the visualization engine, as shown in Fig. 2 (Havard et al. 2019; Ma et al. 2019; Wang et al. 2021). This visualization engine can provide a virtual or augmented reality rendition of the digital twin in the virtual or real worlds which a designer or potential user can interact with to improve insights and understanding until the real-world twin is available.

4 Defining human digital twin and human digital twin systems

Based upon the prior discussion, we can attempt to define a human digital twin and a human digital twin system. We need to consider several aspects to define the human digital twin. It is clear based upon the literature that a human digital twin system (HDTS) should include a digital representation of a real-world human. This real-world human may be an individual, whom we wish to characterize and model or a human class, where this human class represents a group of humans with common traits, characteristics, behavior, etc. The human digital twin (HDT) is then the digital representation of the real-world human or real-world twin. The HDT may exist purely as a mathematical representation of the individual or class of individuals. Alternately, the HDT may exist as a virtual entity which can be rendered within a virtual or real-world system.

The digital representation can include both first principles models which are based on fundamental understanding and statistical models. Various attributes of one or more humans can be modeled (Alexander et al. 2020). These include attributes in at least one of the following categories:

  1. a

    Physical: including anthropometric attributes, biomechanics attributes, times required for task completion, eye movements, and injuries

  2. b

    Physiological: measures such as heart rate, heart rate variability, galvanic skin response, muscle tension, blood oxygen level, brain electrophysiologic signals, pupillometry, blink rates and timing, peripheral blood flow, and gastronomical activity. However, these are intended to correlate to higher level physiological measures such as fatigue, circadian cycle, alertness level, and activation or engagement level.

  3. c

    Perceptual performance: auditory sensitivity, ability to decipher speech in different languages, visual sensitivity, color sensitivity, contrast sensitivity, pressure sensitivity, pain thresholds, temperature sensitivity, etc.

  4. d

    Cognitive performance: knowledge, skills, abilities or aptitudes, workload level, situation awareness, decision-making abilities, intuitive/analytic bias, etc.

  5. e

    Personality characteristics: attributes such as personality type, propensity to trust, and propensity towards suspicion.

  6. f

    Emotional state: feelings experienced by the individual within their current circumstance, including levels of depression, and anxiety.

  7. g

    Ethical stance: representation of the individual’s belief system, including values, beliefs, and mores.

  8. h

    Behavior: actions taken by the individual to interact with the system. These actions may be influenced by any of the earlier attributes.

As such, we can define a human digital twin as an integrated model which facilitates the description, prediction, or visualization of one or more characteristics of a human or class of humans as they perform within a real-world environment. A human digital twin system is a pairing of a real-world human twin and a human digital twin which includes a model of physical, physiological, personality, perception, cognitive performance, emotion, or ethics of a human; where the real-world human and human digital twin are integrated such that a change in the real-world human or its digital representation produces change(s) in the other. This human digital representation likely includes both mechanistic and statistical models of one or more attributes related to one or more of the human defined properties.

5 Human digital twin systems use cases

Besides defining HDTS and investigating their structure, it is also useful to understand how they might be applied. Within the literature, use cases are discussed within the health industry where HDTS support well-being, health care, and development of medical devices (Laamarti et al. 2020). Additionally, use cases are discussed for the design and manufacturing of products, including design of human–machine interaction, as well as the development of product service and use (Ma et al. 2019).

For product development and application, use cases may be considered for different product lifecycle phases; that is, they may be useful during product design, verification and validation, manufacturing, and use. Although HDTS might be useful during any phase of the lifecycle, Fig. 3 and Fig. 4 depict the use cases which apply during product development or acquisition, and operations and sustainment, respectively.

Fig. 3
figure 3

Example product development and acquisition-based use cases

Full size image
Fig. 4
figure 4

Operations and sustainment use cases imagined based upon literature

Full size image

As Fig. 3 shows, during the product development and acquisition life cycle phase, HDTS can be used to aid the design of various portions of the system to be acquired. This might start with simulation of the human operating the product within the system of systems (SoS) to understand operational needs and the benefits of enhancements. As these needs are defined, modifications or new product concepts may be developed and evaluated (Laamarti 2019). In this use case, models which exist within the human digital twin can be applied early in system development to evaluate the utility of various designs and to analyze the impact of these differences through trade space analysis (Tao et al. 2019b). This might include evaluation and understanding of various manpower, personnel, and training options, as well as options regarding the performance or other attributes of the system which impact important emergent behaviors, i.e., “ilities” of the system. Furthermore, these models might consider modifications to the user interface, items which enhance human physical or cognitive performance, and personal protective equipment (PPE) to prevent performance degradation. Perhaps, some models might be applied to understand the tradeoffs in operator or maintenance procedures to aid mission design or to optimize the performance of operators or maintainers within the system. As further indicated in this diagram, these models and HDTS will likely require sharing and collaboration among different communities which interact during product design and acquisition.

It should be noted that early in a development cycle of a novel product, the real-world portion of the HDTS will undergo design, and therefore, the real-world system or representative, trained operators may not exist. Thus, a representative real-world system will not be available. During these portions of the project, the digital portion of the HDTS will be required to operate in an open loop fashion. However, as prototypes of portions of the system are developed and early proxies for system operators and maintainers are used to participate in demonstrations, simulations, or evaluations, data from these events can be used to begin to close the loop from the real-world to the digital models for the system and the humans. As this data is included, the level of model fidelity and user confidence in these models should improve. Besides changes in model fidelity, the level of detail represented by the models is likely to change as one progresses through the system development and acquisition process from early needs development through acquisition. Similar models can also be applied during development of the manufacturing and logistics processes to affect human interaction as the real-world twin is developed.

As the system enters operations and sustainment, the use cases depicted in Fig. 4 become relevant. In this phase, the focus of HDTS shift from focusing on changes to the material components of the system to focusing on changes in the tactics, techniques, and procedures (TTPs) which are employed when applying the system. Similarly, these models may be used to evaluate and improve training or personnel selection methods (Ma et al. 2019). Additionally, the human digital twin may be used to understand readiness of personnel who interact with the systems or understand the errors or circumstances in which errors are made (Tao et al.  2019b). Finally, the HDTS may be applied in real time to determine and allocate taskwork among team members (Ma et al. 2019). The corresponding use cases are shown in Fig. 4.

As shown in Fig. 4, as the HDTS is deployed with the system to the field, we expect that the operators will develop and employ new or revised TTPs which increase the utility and performance of the product (Cox and Szajnfarber 2017). The resulting innovations help to redefine taskwork and improve training. Furthermore, the models themselves may be used to explore changes to taskwork beyond those captured from observing user behavior, as observed in the manufacturing-oriented literature. Additionally, the HDTSs may be employed to analyze and understand human errors, which result in either near misses or mishaps to suggest changes in TTPs, training, or taskwork to improve safety. Furthermore, utilization of personal protective equipment, as well as environmental exposure to information, energy, or matter (e.g., radiation or known carcinogens), may be monitored to support occupational health interventions. Furthermore, the HDT may be capable of understanding attributes of the humans which enhance performance, aiding refinement of personnel selection.

Although TTPs, training, and personnel selection may all improve system performance, perhaps the greatest opportunity is understanding human readiness. However, this use case likely diverges significantly from the human digital twins that would be useful during system development. In fact, this use case likely relies upon significant interaction with the management and medical personnel who are not as likely to interact with the other use cases illustrated in Fig. 4. Although not apparent from our earlier discussion, the HDT applied in medical and sports performance focus on understanding performance of systems which are internal to the human while HDT applied in product design or manufacturing tend to focus on human behavior and the interaction of the human with systems and the environment. As such, there is a basis for arguing that the HDT used for understanding readiness may be unique from the HDT used for practically all other use cases shown in Fig. 3 and Fig. 4. Nevertheless, each of these HDTs relies on input from the same real-world humans and the HDT models within these two areas certainly interact. Therefore, a HDT developed to support readiness and HDT developed for other use cases will likely require some level of integration. Thus, there will be a need to integrate models of the human which are intended to understand health or health related human performance with models of the human which are intended to understand their performance when interacting with systems within an environment.

6 Summary and conclusion

This paper sought to define a common structure, definition, and set of use cases for HDTS as they pertain to systems, generally, and to product development, specifically. As discussed, the HDTS terminology has evolved in the past few years and is discussed predominantly in research papers related to medical and manufacturing systems, although other applications including fitness, product design, and military, among others, are present within this literature.

This research provides a definition, structure, and use cases for these systems, as defined within the reviewed literature. However, the development of HDTS as defined in this paper are relatively new, and therefore, the definitions, structure, and use cases are likely to evolve as these systems are employed and used. While an effort was made to capture the structure and behavior of these systems within SysML models and then to integrate these models, these activities required interpretation by the authors. Therefore, these models may not always represent the intent of the authors of the original articles. For each of these reasons, it is hoped that the definition, structure, and use cases of HDTS provided in this paper will provide an integrated view from which these definitions can evolve.

As shown, the HDTS include a real-world twin, a digital twin, and an interchange component which facilitates real-time monitoring of the real-world human as they perform in an environment using the system or product to provide robust models and predictions which facilitate system improvements. HDTS rely upon recent advances in low-cost sensors, data storage, and data mining technologies, together with the ongoing conversion to digital engineering as components. The integration of these components facilitates feedback loops which support improvements both in the design of products or systems for improved performance in real-world settings as well as improvements in TTPs and training to enhance user performance during product or system operations and sustainment.

Successful implementation of HDTS will require the development of robust models of various aspects of the human at different scales for different applications. These models will need to address aspects of the human to include physical, physiological, perceptual, cognitive, behavioral, emotional, personality, and ethical structure and performance. These models will need to describe and account for the variability among and uniqueness of individuals to be suitable for application within HDTS. The challenge, however, is the fact that component models which address, typically narrowly defined, aspects of the human are being developed within several disciplines for disparate applications as illustrated by the literature review. To accelerate the deployment of this technology, a concerted, multi-disciplinary, effort will be required to understand, define, and synthesize these models into HDTS applications.