systems

Article
MBSE with/out Simulation: State of the Art
and Way Forward
Bernard P. Zeigler 1, *, Saurabh Mittal 2 and Mamadou Kaba Traore 3
1 Co-Director of the Arizona Center for Integrative Modeling and Simulation (ACIMS),
University of Arizona and Chief Scientist, RTSync Corp., 6909 W. Ray Road, Chandler, AZ 85226, USA
2 MITRE Corporation, 7515 Colshire Dr., McLean, VA 22102, USA; smittal@mitre.org
3 IMS CNRS UMR 5218, University of Bordeaux, 33405 Talence, France; traore@isima.fr
* Correspondence: zeigler@rtsync.com




Received: 11 October 2018; Accepted: 9 November 2018; Published: 15 November 2018


Abstract: The limitations of model-based support for engineering complex systems include limited
capability to develop multifaceted models as well as their analysis with robust reliable simulation
engines. Lack of such Modeling and Simulation (M&S) infrastructure leads to knowledge gaps in
engineering such complex systems and these gaps appear as epistemological emergent behaviors.
In response, an initiative is underway to bring Model-Based Systems Engineering (MBSE) closer
together with model-based simulation developments. M&S represents a core capability and is
needed to address today’s complex, adaptive, systems of systems engineering challenges. This paper
considers the problems raised by MBSE taken as a modeling activity without the support of
full strength integrated simulation capability and the potential for, and possible forms of, closer
integration between the two streams. An example of a system engineering application, an unmanned
vehicle fleet providing emergency ambulance service, is examined as an application of the kind of
multifaceted M&S methodology required to effectively deal with such systems.

Keywords: modeling and simulation; discrete event system specification; DEVS; model-based system
engineering; MBSE; internet of things; IoT; cyber physical systems; CPS; complex adaptive systems
of systems

1. Introduction
Model-based engineering originated in the 1970s and with the foundational Systems Theory
providing means and methods to incorporate simulation as integral mechanism to understand
the abstractions and conceptual alignment between various constituent parts/systems. A. Wayne
Wymore’s book [1] is generally acknowledged as the first formulation of Model-based System
Engineering (MBSE). Quite fittingly, Wymore is also one of the early System Theorists and the theory
in his book [2] proves the basis for the Discrete Event System Specification (DEVS) to be discussed
here. With the advent of Information Technology in the late 1990s, new modeling notations emerged
that helped develop IT-enabled systems using traditional systems engineering practices. With IT
now woven in every fabric of society, IT-enabled systems have grown complex and unmanageable.
These are commonly known as sociotechnical systems [3].
To describe this new class of super complex systems in a man-made world, labels such as
System of Systems (SoS), Cyber-Physical Systems (CPS), Complex Adaptive Systems (CAS), and Cyber
CAS (CyCAS) are used interchangeably [4,5]. All of them are multi-agent systems, i.e., have large
number of agents, are contextualized in an interactive environment and manifest emergent behavior.
The constituting agents are goal-oriented with incomplete information at any given moment and
interact among themselves and with the environment. SoS is characterized by the constituent

Systems 2018, 6, 40; doi:10.3390/systems6040040 www.mdpi.com/journal/systems

Systems 2018, 6, 40 2 of 18

systems under independent operational and managerial control, geographical separation between the
constituent systems and independent evolutionary roadmap. CAS is an SoS where constituent systems
can be construed as agents that interact and adapt to the dynamic environment. Cyber CAS is a CAS
that exist in a netcentric environment (for example, Internet) that incorporates human elements where
distributed communication between the systems and various elements is facilitated by agreed upon
standards and protocols. CPS is an SoS wherein the constituent physical and embedded systems are
remotely controlled through the constituent cyber components.
Model-based Systems Engineering employs model-based practices to engineer IT-enabled systems.
While they still can be created using MBSE practices, the usage of such systems is far from it. In the
sociotechnical era of Internet of Things (IoT), wherein multiple domains (for example, cyber, physical,
and computational across various societal sectors) are involved, experimenting with the model to
understand the model’s functionality and engineer the resulting complex system is a challenging task.
The existing toolsets lack the needed simulation analysis and experimentation capabilities leading
to epistemological emergent behaviors, which is a characteristic defining property of any complex
system [6]. These emergent behaviors can be both positive and negative. The negative emergent
behaviors lead to cascaded failures while positive emergent behaviors may be sustainable and improve
the complex systems’ overall function [7]. To overcome such limitations requires extending MBSE for
human machine interaction analysis and resilient system design [8].
The model-based paradigm has been successfully adopted by many disciplines when it comes
to traditional systems engineering. However, for complex systems engineering, it is still in infancy
as the tool-set support for engineering complex systems is limited [9]. These limitations range from
developing multifaceted models (including comprehensive, highly detailed simulations and highly
informative, analytic simplifications) to providing equally capable underlying simulation engines.
Lack of such M&S infrastructure leads to knowledge gaps in engineering such complex systems and
these gaps appear as epistemological emergent behaviors [10]. Many domains and disciplines are
successfully employing simulation capabilities within their domains [11]. Furthermore, strong support
for developing enterprise system of systems using architecture frameworks [12] such as Department
of Defense Architecture Framework (DoDAF), Unified Architecture Framework (UAF), etc. is now
widely available. However, these are nowhere close to addressing the requirements for multifaceted
model development and its underlying simulation and experimentation infrastructure.
For the purpose of a unifying label, we will consider the term Complex Adaptive System of
Systems (CASOS). This term emphasizes three aspects: complex, adaptive and SoS of complex
sociotechnical systems. These three are distinct characteristics that require unique infrastructure (both
hardware and software) for implementation. An SoS may be complicated but not complex i.e., SoS
may employ traditional systems engineering. The adaptive aspect is brought in the mix through
agent embodiment, situatedness and learning. These are made available in sociotechnical systems
through Artificial Intelligence, Machine learning, algorithms, or the presence of human-in-the-loop
(that complement the system functionality with guidance and participatory roles). We consider a class
of examples of such CASOS in Appendix A to illustrate the particular advances in M&S required to
support their systems engineering design.
CASOS present challenges that cannot be easily tackled using MBSE nor classical modeling,
simulation, and optimization techniques. Recent model-based system engineering has proved
inadequate due to lack of a full-strength M&S computational substrate [5]. Modeling and Simulation
(M&S) methodology has been evolving to provide increasing capability to help systems engineers
develop models of CASOS [4,11,13]. Such simulation models support design and testing of mechanisms
with learning capabilities to coordinate the interactions of the operationally and managerially
independent components. The design of such systems presents challenges to the currently employed
independent use of simplified models for formal verification or brute-force simulations which are
severely limited in the range of conditions they can test. M&S of CASOS must have a usable modeling
environment that facilitates model validation from the end-user and a robust simulation infrastructure
Systems 2018, 6, 40 3 of 18

that can be formally verified to ensure correct model execution. Together, they enable exhaustive
parameter evaluation and advanced experimentation. Model-based methods which support traditional
systems engineering need to be augmented with simulation-based methodologies to ensure they
support complex systems engineering that integrate discrete and continuous systems for complex
hybrid systems. CASOS engineering will not become possible unless undesired emergent behaviors
are completely removed from a computational environment or are known a priori so that they can be
knowledgeably eliminated. A computational simulation-based environment provides experimentation
opportunities to validate a CASOS model, such that it becomes predictable and eventually useful [5].
Ultimately, this is realizable in a Live, Virtual and Constructive environment with robust simulation
infrastructure and human-in-the-loop undertakings [14].
The task of integrating various simulators to perform together as a composite simulation, termed
as co-simulation, involves weaving the time series behavior and data exchanges accurately, the failure
of which will yield inaccurate simulation results. As elaborated by Mittal and Zeigler [15], in the
absence of a generic approach, every such hybrid system would require a dedicated effort to build a
co-simulation environment. Bringing various simulators together is much more than a typical software
engineering integration exercise.
In the following sections, we start with some background in M&S theory and the Discrete Event
System Specification (DEVS) formalism to lay the basis for discussion of multifaceted modeling and
the associated co-simulation infrastructure. This sets the table for considering how MBSE, DEVS
and CASOS may be unified. Then, a generic architecture and workflow are proposed for M&S
working within MBSE. This leads to a more in-depth discussion of multiobjective, multiperspective,
and multiresolution families of models supporting simulation capabilities and tools. An example
of a system engineering application, an autonomous unmanned vehicle fleet providing emergency
ambulance service, is examined as an application of the kind of multifaceted M&S methodology
required to effectively deal with such systems.
The paper is organized as follows: Section 2 describes the developments in M&S using DEVS
formalism and its application to Complex Adaptive Systems, Section 3 describes the concepts in
multifaceted M&S that facilitate development of multiresolution, multiperspective and multiobjective
modeling and simulation. Section 4 provides an overview of the State of the Art of simulation
tools. Section 5 discusses the way forward for both MBSE with, and without, simulation, followed by
conclusions in Section 6.

2. Developments in Modeling and Simulation and DEVS

A solution to the problem of developing models and associated simulation capabilities that is
gaining increased acceptance is offered by the DEVS formalism with a holistic construct called the
Modeling and Simulation Framework (MSF). Briefly summarized, the framework defines the entities
and their relationships of the enterprise of M&S and includes the relation between detailed models
and their abstractions [16]. The framework is based on mathematical systems theory and recognizes
that the complexity of a model can be measured objectively by its resource usage in time and space
relative to a particular simulator, or class of simulators. Furthermore, properties intrinsic to the model
are often strongly correlated with complexity independently of the underlying simulator. Successful
modeling can then be seen as valid simplification, i.e., reduction of complexity to enable a model to be
executed on resource-limited simulators and, at the same time, creating morphisms that preserve behavior
and/or structural properties, at some level of resolution, and within some experimental frame of interest. Indeed,
according to the framework, there is always at least one pair of models involved, which are called the
base and lumped models, in such a relationship.
The DEVS formalism is formulated within MSF and formally specifies the internal behavior
of the system as well as macro behavior of the overall system due to its closure under coupling
property. This robustness in both structural and behavioral description ensures that the unwanted
holistic behaviors, also known as negative emergent behaviors are explicitly avoided, along with
Systems 2018, 6, 40 4 of 18

the guaranteed manifestation of the desired (or positive) emergent behaviors [6,13,17]. The DEVS
super-formalism provides a foundation [15] that specifies an abstract simulation protocol between
the model and the simulator [16]. Thus, a requirement for M&S of CAS is to employ the principles
of the Parallel DEVS simulation protocol (as illustrated by the hybrid approach of Camus et al. [18],
for example) to support the required robust co-simulation.

2.1. MBSE, DEVS and CAS: Towards Unification

A major thrust of the MBSE community calls for formalized models to replace documents
as the fundamental building blocks of systems engineering [19] Such models support a host of
model-intensive activities such as improved communication about system artifacts among stakeholders
as well as strengthened testing and verification (both at design as well as at runtime). Practical
implementation of this thrust demands that such models eventually support all the activities typically
associated with the simulation discipline. However, as suggested, current MBSE formalisms stop
well short of this capability. One approach to bridging this gap is to enable mappings to be defined
that precisely specify simulation models that realize the models’ behaviors. Taken to a logical limit,
this approach entails building more capability into a MBSE formalism so that it eventually replicates all
capabilities associated with traditional simulation methodology. Although there are attempts to achieve
this goal [20,21], there are also fundamental reasons why it is not attainable [22,23]. An approach that
we hypothesize to work is to tie MBSE models with informal but well documented links. Furthermore,
as experience grows with such cross-linkages, it might eventually become feasible to formalize these
associations. Another complimentary approach is model-based interactive storytelling (MBIS) that
enhances MBSE and interactive storytelling to increase stakeholder participation in the systems
engineering process especially involving participants from multiple disciplines, and eventually
transdisciplines (see below) [24].

2.2. Architecture and Workflow for M&S Working within MBSE

Software defined systems must increasingly operate on large, time-varying, heterogeneous data.
Big Data enables and requires that these systems perform across an enormous variety of operating
conditions presenting engineers with multi-dimensional, hierarchical, uncertain and critical control
and decision challenges requiring transdisciplinary systems engineering [24]. Recent work has begun
to address these challenges. Kavak et al. [25] offer a structured modeling approach to produce agents
or parts thereof directly from data that focuses on individual-level data to generate agent behavioral
rules and parameter values. Generalizing from the approach that recently enabled the AlphaGo
program to defeat the world’s top ranked human Go player, Wang et al. [26] envision an AlphaGo-like
computation platform to enable artificial systems to model and evaluate complex systems, and through
the virtual-real system interaction, realize effective control and management over the complex systems.
The architecture of Figure 1 offers a generic workflow that supports Wang et al. [26] vision of
AlphaGo-like computational strength for future M&S-based systems engineering and management.
The Modeling and Simulation Framework [16], and, in particular, its system specification hierarchy
for acquiring levels of knowledge about an observed system, provides a solid basis for inference of
structures from the volumes of Big Data envisioned by Kavak et al. [25].
Figure 1’s architecture and workflow for M&S offers a vision of model production for use in MBSE.
The process starts with the development (or reuse) of a System Entity Structure (SES) that organizes a
family of simulation models for the current application of interest [27]. SES is an ontology, a language
with syntax and semantics to represent declarative knowledge [28]. The SES representation scheme
structures the search for a subset of models that are of particular concern under criteria that relate
essentially to their behavior and can’t be defined in the first instance by their structural properties.
Indeed, the behavior generated under simulation is observed within the Experimental Frame that
characterizes the criteria defining the subset of interest. Roughly, an experimental frame (EF), as defined
within the MSF, is a specification of the conditions under which the system is observed or experimented
Systems 2018, 6, 40 5 of 18

with. As such, experimental frames are the operational formulation of the criteria that motivate the
Systems 2016, 4,pursuit
M&S-based x FOR PEER of theREVIEW
models of interest. The SES includes coupling information that directs5 of 20
the
compositions of hierarchical models from components in the model-base. The combination of selection
combinatorial
from specializations search andspace. Since
aspects leadsantoSES describes
a very a number of search
high combinatorial systemspace.
configurations,
Since an SES thedescribes
SES tree
needs to be pruned to get one particular configuration, which is called
a number of system configurations, the SES tree needs to be pruned to get one particular configuration, Pruned Entity Structure (PES).
Pruning
which operations
is called Pruned factor outStructure
Entity a particular (PES).model
Pruning specification
operations which
factorcan outthen be transformed
a particular model
automatically into a coupled model with components from the
specification which can then be transformed automatically into a coupled model with components model base. Such components are
eitherthe
from DEVSmodel modelsbase.orSuchhavecomponents
been wrapped areineither
a DEVS DEVSinterface
models for or
DEVShavecompliance
been wrapped and amenability
in a DEVS
to the coupling specified by the SES. Simulation of such a model, eventually
interface for DEVS compliance and amenability to the coupling specified by the SES. Simulation of such on a high-performance
aplatform using parallel
model, eventually on asimulations
high-performanceof multiple models
platform under
using testingsimulations
parallel for reasonable execution
of multiple times,
models
generates
under testingtheforbehavior
reasonable of the model and
execution produces
times, generatesresults
the in the experimental
behavior of the model frameand of interest.results
produces These
results measure the extent to which the governing criteria are satisfied
in the experimental frame of interest. These results measure the extent to which the governing criteria and are analyzed for guidance
to direct
are theand
satisfied pruning towardsfor
are analyzed a larger percentage
guidance to directofthemodels
pruning thattowards
fully satisfy the percentage
a larger criteria. At of this point,
models
Artificial
that fully Intelligence (AI) is useful
satisfy the criteria. At this to point,
help analyze
ArtificialtheIntelligence
results and predict which to
(AI) is useful new prunings
help analyze ofthe
the
SES should performed at the next iteration. Built into the iteration loop
results and predict which new prunings of the SES should performed at the next iteration. Built into is a second cycle of transition
between
the baseloop
iteration and is lumped
a secondmodels
cycle where the lumped
of transition modelbase
between canandgreatly accelerate
lumped modelsthe search
where thefor high-
lumped
value models by enabling faster runs that provide useful information
model can greatly accelerate the search for high-value models by enabling faster runs that provide for the more detailed base
model. Some fundamental distinctions between base and lumped
useful information for the more detailed base model. Some fundamental distinctions between base and models concern objectives,
representation,
lumped models concern entity objectives,
attributes representation,
and variables,entity interaction
attributes processes, timing
and variables, mechanisms,
interaction and
processes,
computational complexity [16].
timing mechanisms, and computational complexity [16].
Asillustrated
As illustratedininFigure Figure 1, the
1, the architecture
architecture envisions
envisions a collaboration
a collaboration between between
humanhumanand AI and AI
agents.
agents.
The The modeler
human human modeler develops develops
the SESthe andSEStheand
DEVS the DEVS
modelmodel base tobase
span to configuration
span configuration spacespace
that
that encompasses the subset of interest. The AI agent, under control
encompasses the subset of interest. The AI agent, under control of the user, analyses the results of the user, analyses the results
and
generates new prunings in order to increase the percentage of models of interest. Here, we lean onlean
and generates new prunings in order to increase the percentage of models of interest. Here, we the
on thetoagent
agent provide to provide
the gristthe forgrist for patterns
patterns that humans
that humans can discerncan discern and exploit
and exploit with imagination,
with imagination, novel
novel insights,
insights, and storytelling
and storytelling abilityability [24].modeler
[24]. The The modelerdevelopsdevelops
validvalid simplification
simplification morphisms
morphisms for
for the
the DEVS
DEVS base and baselumped
and lumpedmodelsmodelsand decidesand decides
when and when howand how to
to iterate iterate the
between between
levels oftheresolution
levels of
in order to accelerate the overall process [29]. Here, a base-lumped pair of models refers to arefers
resolution in order to accelerate the overall process [29]. Here, a base-lumped pair of models pair ofto
a pair of models—the first more “complex” than the second, which
models—the first more “complex” than the second, which are equivalent in an experimental frame of are equivalent in an experimental
frame ofininterest
interest the system in the system investigation
investigation [16]. Such[16]. Such equivalence
equivalence allows theallows
lumpedthe lumped
model modelintofor
to stand stand
the
base model for the objectives underlying the frame. Thus, while a lumped model is not necessarilynot
in for the base model for the objectives underlying the frame. Thus, while a lumped model is a
necessarily a uni-directional morphic projection of a base model, formalization
uni-directional morphic projection of a base model, formalization and tool support of model-to-model and tool support of
model-to-modelistransformations
transformations a promising technology is a promising
to supporttechnology
the workflowto support the workflow
illustrated in Figureillustrated
1 [30]. in
Figure 1 [30].
To plumb these concepts in more depth, we turn towards multiresolution modeling methodology
in theTo plumb
broader theseof concepts
context multifaceted in modeling
more depth, we turn[27].
methodology towards multiresolution modeling
methodology in the broader context of multifaceted modeling methodology [27].

Figure 1. Architecture and Workflow for M&S working within MBSE.

Figure 1. Architecture and Workflow for M&S working within MBSE.

3. Multifacetted Modeling and Simulation: Multiobjective, Multiperspective, Multiresolution

Families of Models
Systems 2018, 6, 40 6 of 18

3. Multifacetted Modeling and Simulation: Multiobjective, Multiperspective, Multiresolution

Families of Models
Modeling and simulation are activities undertaken to support system engineering decision
making—the ability to assess the effects of constructions and interventions before they are actually
carried out and to pre-select promising ones, in view of the driving objectives. These objectives in turn
serve to orient modeling efforts. We recognize that a CASOS may be subject to a multiplicity of system
engineering objectives.

3.1. Multi-Perspective Families of Models

Multiple objectives require different levels of explanation be provided for the same system under
study. Each level of explanation can be expressed within a dedicated EF, the one that provides answers
to questions of interest from the perspective of its corresponding objective. Therefore, a family of
perspective-specific models needs to be built.
While, in practice, each of these partial models is executed in isolation to provide a given level
of explanation for the system of interest, all of them are related in reality, since they depict various
abstractions of the same system, but from different perspectives (or facets). Therefore, while each
perspective-specific model produces results within its corresponding experimental frame, partial
models can be holistically integrated within a holistic experimental frame, and questions that are
transversal to different perspectives can be accurately addressed, which is not possible in any of the
perspective taken alone.
How do we identify the facets/perspectives in a general context and in a systematic way? We review a
structured approach to building an ontology for the M&S of a domain of interest. The domain analysis
ontology must provide a formal way to capture all the knowledge that might be in the range of M&S
of the domain for which it is likely to be used. Therefore, it must capitalize on the abstractions used
for the simulation of the entire targeted domain, beyond aspect specific modeling. Thus, the generic
approach to domain analysis is a 4-layered ontology which highlights at each layer a generic key
characteristic. As depicted by the SES presented in Figure 2, the following layers are defined:

• System level, where meaningful specializations of the class of systems that characterizes the
domain of interest are highlighted,
• Facet level, where all cumulative aspects of a domain system are clearly separated,
• Scale level, where major spatial and temporal scales are emphasized, Model level, where conventional
models often originating from decades of theoretical findings are identified as reusable artefacts to
be selected and integrated in new studies.

The System level recognizes the whole complex system as a juxtaposition of multiple facets,
while various specializations can be identified as possible instances of the same integrated set of
facets in various specific contexts. For example, healthcare systems can be specialized into primary,
secondary, ternary, and home care [31], while transportation systems can be specialized into air, ground,
rail, and aquatic transport, and military systems can be specialized into air, ground, and marine forces.
The Facet level establishes three generic facets, i.e., “production facet”, “consumption facet” and
“coordination facet”. Although the identification of a system’s facets may depend on the domain
as well as the experts involved and the objectives in mind, we suggest the systematic adoption of
these generic patterns. In other words, a complex system is made up of one or various facets, each of
which being a production system (hence, leading to a ProF model), a consumption system (that gives a
ConF model), or a coordinating system between production and consumption (giving a CooF model).
These patterns encompass the traditional supply-demand duality that often characterizes complex
systems [31]. The notion of “Production” encompasses the notion of “Supply” in that it involves not
only the intentional supply of services needed, but all phenomena that produce positive and negative
impacts on the system’s stakeholders. Examples of production in healthcare include vaccination and
information diffusion (as production of ease), but also contamination and epidemics (as production of
Systems 2018, 6, 40 7 of 18

disease).
Systems 2016,Examples in transportation
4, x FOR PEER REVIEW include the production of public or private transportation services, 7 of 20
but also the production of air pollution, land use, and accidents. Examples in military systems include
infrastructure
the production of destruction.
security andSimilarly,
protection,the butnotion
also theofproduction
“Consumption” encompasses
of life and infrastructure thedestruction.
notion of
“Demand”, as consumers may not be only users seeking intentionally
Similarly, the notion of “Consumption” encompasses the notion of “Demand”, as consumers may not for services but all stakeholders
that
be consume
only what is
users seeking produced by
intentionally for the system.
services but Examples
all stakeholdersof consumers
that consume are population,
what is produced patients,
by
travelers, pedestrians, territories, enemies, etc. An important element
the system. Examples of consumers are population, patients, travelers, pedestrians, territories, enemies, of this multi-perspective
approach
etc. is that, while
An important elementperspectives have mutual influence
of this multi-perspective approachon is each
that, other, each perspective
while perspectives havecaptures
mutual
its received influences by means of parameters, which values explicitly
influence on each other, each perspective captures its received influences by means of parameters, reflect implicit assumptions
and simplifications
which values explicitly made about
reflect otherassumptions
implicit perspectives and influences.
simplificationsFor example,
made about when focusing
other on the
perspectives
system as aFor
influences. production
example,system, the ProFon
when focusing model will make
the system as ause of parameters
production system, (such
theasProFthe model
arrivalwill
rate
of patients in a hospital, or the arrival rate of travelers in a shuttle, or the
make use of parameters (such as the arrival rate of patients in a hospital, or the arrival rate of travelersarrival rate of enemies in a
combat theatre) to aggregate all processes going on in the same system
in a shuttle, or the arrival rate of enemies in a combat theatre) to aggregate all processes going on when taken as a Consumption
system.
in the same In its turn,when
system the ConF
takenmodel will make use
as a Consumption of parameters
system. In its turn,(such as themodel
the ConF deathwill ratemake
of a
population) to aggregate all processes going on in the same system when
use of parameters (such as the death rate of a population) to aggregate all processes going on in the considered as a Production
system.
same systemCoordination can beasseen
when considered as cross-organization
a Production system. Coordination mechanism can bemanaging the entities and
seen as cross-organization
resources ofmanaging
mechanism existing ones, such that
the entities andindividual
resources of goals as well
existing ones,as such
system-wide goals are
that individual satisfied.
goals as wellItasis
needed to the extent that specific interaction and cooperation are required
system-wide goals are satisfied. It is needed to the extent that specific interaction and cooperation are to ensure safe entangled
and context-dependent
required behaviors.and context-dependent behaviors.
to ensure safe entangled
The Scale level emphasizes
The Scale level emphasizes on onthat
thataacharacteristic
characteristic feature
feature of of complex
complex systems
systems is is the
the occurrence
occurrence
of interactions between heterogeneous components at different
of interactions between heterogeneous components at different spatial and temporal scales with spatial and temporal scales with
variousinterpretations
various interpretationsofof thethe notion
notion of scale,
of scale, andand a major
a major concern concern
aboutabout scale transfer
scale transfer processes processes
where
inter-scale interactions must be properly described, as emphasized in [31]. A scale refers to a settoofa
where inter-scale interactions must be properly described, as emphasized in [31]. A scale refers
set of relationships,
relationships, whichwhich implicitly
implicitly (or explicitly)
(or explicitly) point point to spatial
to spatial properties
properties (such(such as location,
as location, shape, shape,
size,
size, etc.), as well as temporal properties (such as exact or approximate
etc.), as well as temporal properties (such as exact or approximate timing, simultaneity or sequentiality, timing, simultaneity or
sequentiality, continuity, etc.). Thresholds between scales are critical
continuity, etc.). Thresholds between scales are critical points along the scale continuum where a shift points along the scale
continuum
in the importance whereofa variables
shift in the importance
influencing of variables
a process occurs. influencing
As a result, a process
the generic occurs. As a result,
ontology proposedthe
generic ontology proposed exhibits macro, meso and micro levels
exhibits macro, meso and micro levels of abstraction both within the consumption and the production of abstraction both within the
consumption and the production facets, leading respectively
facets, leading respectively to the generic MaConF, MeConF, MiConF, MaProF, MeProF and MiProF to the generic MaConF, MeConF,
MiConF,The
models. MaProF, MeProF and MiProF
macro-meso-micro models.
architecture The macro-meso-micro
is recognized as describingarchitecture is recognized
the three possible as
levels of
describing the three possible levels of inquiry on
inquiry on which social scientific investigations might be based [32]. which social scientific investigations might be based
[32].

Figure 2. Generic SES for Multi-perspective Modeling.

Figure 2. Generic SES for Multi-perspective Modeling.

The Model level identifies conventional models often originating from decades of theoretical
findings as reusable artefacts to be selected and integrated in new M&S studies of complex systems.
It defines the abstractions that can be directly simulated, by distinguishing four generic types of
model, i.e., entity models, flow models, functional models and spatial models. While Entity models
Systems 2018, 6, 40 8 of 18

The Model level identifies conventional models often originating from decades of theoretical
findings as reusable artefacts to be selected and integrated in new M&S studies of complex systems.
It defines the abstractions that can be directly simulated, by distinguishing four generic types of model,
i.e., entity models, flow models, functional models and spatial models. While Entity models describe
autonomous individuals with specific attributes and with or without goal-driven behavior, functional
models are formulated as mathematical equations, spatial models are composed of individuals
geographically located in a space model, and flow models capture scenarios an individual can
undergo. These models explicitly describe the temporal and spatial properties pointed out (implicitly
or explicitly) at the scale level. Consequently, the generic ontology has in each facet, entity and flow
models at the micro level of abstraction, and functional and spatial models at the macro and meso
levels of abstraction. It is worth noting that the fact that a spatial model at any macro level involves
a space model that contains abstractions detailed at lower levels (i.e., meso and micro), and that,
similarly, a spatial model at any meso level involves a space model that contains abstractions detailed
at the micro level.
The generic ontology is meant to be instantiated in the analysis of any new domain of interest
in view of its M&S. Such an instantiation provides the domain-specific ontology that will drive the
multi-perspective modeling and holistic simulation (MPM&HS) process of the targeted domain.

3.2. Multi-Resolution Families of Models

Multi-resolution modeling (MRM) is essential for exploratory analysis of CASOS design spaces
because it is neither cognitively nor computationally possible to keep track of all relevant variables
and causal relationships [33]. A typical multiresolution scenario applicable to defense investigates the
operational differences between low-level military entities such as individual tanks and the aggregated
high-level units, e.g., battalions or platoons when moving in a battlefield. Attributes of an aggregated
entity like a tank battalion are often determined by applying an aggregation mapping to the attributes
of its individual entities. The mapping can group a set of tanks to a single tank battalion together
with a function to derive holistic attribute values, e.g., an average speed of a tank battalion, from the
constituent individual tank speeds (disaggregation is the inverse mapping). Here, the base model
is typically “more capable” and requires more resources for interpretation and simulation than the
lumped model. By the term “more capable”, we mean that the base model is valid within a larger set
of experimental frames (with respect to a real system) than the lumped model. Here, we note that the
terms “base” and “lumped” are terms employed with the framework to denote the full range of possible
pairs of models in which the first is more capable (e.g., more detailed, disaggregated, high resolution,
fine grained) than the second (less detailed, aggregated, low resolution, coarse grained). We note that
MRM sometimes refers to simulation environments in which entities are aggregated or disaggregated
at runtime based on certain triggering conditions, in which case, resolution changes dynamically at
runtime. Such dynamic structure models are included within the larger category of families of models
structured at different resolutions meeting different modeling objectives.
The scope/resolution/interaction product is limited by the computational resources at hand.
Thus, typically a large scope places constraint on the resolution and interaction that can be represented,
whereas a smaller scope will permit higher resolution and interaction. Nevertheless, the trend towards
high fidelity M&S implies increasing all factors in the scope/resolution/product (as in the base model)
while lumping can reduce both scope and/or resolution independently.
Some typical distinctions often drawn between base and lumped models with respect to agent
modeling are presented in Table 1 [33,34].
Systems 2018, 6, 40 9 of 18

Table 1. Some fundamental distinctions between base and lumped models (in a military simulation example).

Base Model Lumped Model

Results traceable to specific performance
data and assumptions. Predict overall results
Evaluate subtle differences in weapons, Include small numbers of parameters
Objectives
sensors, or tactics, Parameter values amenable to identification
Understand how different inputs affect from feasibly obtainable data
combat performance
Aggregate entities into groups typically
Representation Individual agents as separate entities
respecting command hierarchy.
Location in space and time, position in Averaged entity values attributed to groups,
social or other hierarchies, perception of Discrete events compounded into rates
Entity Attributes
the situation: threats and opportunities, for groups,
and Variables
capabilities, etc. updated at event Global state sets, cross-products of individual
occurrences or time steps state sets
Decomposed into sequences of events
Processes aggregated into group level
Interaction processes and activities,
formulae abstracting individual behavior
Tracking of individual behaviors
Micro stochastic sequences can be aggregated
Coordinate the event sequences for the
into macro behaviors using law of large
Timing mechanisms numerous participants so that subtle
numbers expressed more simply in stochastic
interaction patterns can be modeled
or deterministic form
Computational
Complexity Lean towards large scope, high resolution Lean towards smaller scope, low resolution
(Scope/resolution/ and unconstrained interaction and constrained interaction
interaction product)

However, the important point is that within a particular experimental frame of interest the lumped
model might be just as valid as the base model. Furthermore, the trade-off between performance and
accuracy [35] is a fundamental consideration where performance refers to the computational resources
used in a simulation run and accuracy refers to the validity of a model with respect to a referent system
within an experimental frame [16]. Use of computational resources tied to a simulator’s time and space
demands in generating the model’s behavior are correlated with its scope/resolution product where
scope refers to how much of the real world is represented, resolution refers to the number of variables
in the model and their precision/granularity.
Models should be built in an incremental manner with continuous engagement and validation
from the subject matter experts (SMEs) and their mapping to the experimental frames. This pairing
allows the selection of the correct resolution of the model. This allows the development of early insight
into the objective of the modeling, provides a holistic view of the system under study, provides a
testing framework for the target, explains the target’s behavior early in the M&S development cycle
and serves as the foundation to add more complexity to the subsequent models. Parallel development
of the experimental frames provides the experimentation and data collection requirements for the
model and the computational requirements of the underlying simulation platform.
The incremental manner from lower complexity to higher complexity works when one tries
to build models from top-down. However, there may be situations that in high fidelity models are
already present and one needs to bring them together to develop a larger SoS model. This includes
the incorporation of legacy models and simulators. In that situation, lumping needs to be carefully
managed as information is lost in aggregation if not justified by valid abstractions such as from the
underlying fundamental science (e.g., physics, chemistry, etc.). This is then further supported by
systems morphisms and homomorphisms to ensure that there is continued correspondence between
the model family (containing base, lumped and higher fidelity) and the model hierarchy. Figure 3
provides an illustration of the construction of such a model family as built up from base/lumped model
pairs. For example, a base model may be composed of multiple components each of which can be
lumped into simplified lumps and coupled together constitute a new lumped model. The base model
 underlying fundamental science (e.g., physics, chemistry, etc.). This is then further supported by
systems morphisms and homomorphisms to ensure that there is continued correspondence between
the model family (containing base, lumped and higher fidelity) and the model hierarchy. Figure 3
provides an illustration of the construction of such a model family as built up from base/lumped model
pairs.2018,
Systems For 6,example,
40 a base model may be composed of multiple components each of which 10 can be
of 18
lumped into simplified lumps and coupled together constitute a new lumped model. The base model
might itself serve as a lumped component of a larger model leading to a hierarchical construction.
might itself aserve
Moreover, lumped a lumped
as model maycomponent
itself serveofasa alarger modelfor
base model leading to a hierarchical
an abstraction construction.
that supports reduced
Moreover, a lumped model may itself serve as a base
resolution. Thus, working together, resolution and composition operations can create reduced
model for an abstraction that supports a multi-
resolution.
resolutionThus,
familyworking together,
of simulation resolution and composition operations can create a multi-resolution
models.
family of simulation models.

Figure 3. Relationship between the base and the lumped models within the model family.
Figure 3. Relationship between the base and the lumped models within the model family.
A methodology for constructing a multiresolution family of models is illustrated in Figure 4,
A methodology
as follows: for constructing
given requirements a multiresolution
and constraints familyconsider
of the problem, of models is illustrated
a model in Figure
that satisfies 4, as
all such
follows: given requirements and constraints of the problem, consider a model
requirements and constraints as the base model to be aimed for. Create lumped models by making that satisfies all such
requirementsabout
assumptions and constraints as the
the base model, base model
including to beofaimed
relaxing for. Create
constraints lumped models
and dropping by making
of requirements.
assumptions about the base model, including relaxing of constraints and dropping
Create higher resolution models by removing assumptions that were previously added while including of requirements.
Create
more higher
refined resolution models
representations by the
to address removing assumptions
affected constraints andthat were previously
requirements, meanwhileadded while
checking
including
for consistencymore refined representations
of predictions between related to base
address the affected
and lumped models.constraints
The targetedand baserequirements,
model is the
Systems 2016, 4, x FOR PEER REVIEW 11 of 20
meanwhile
one achievedchecking
when all for consistency
assumptions thatofhave
predictions
been madebetween relatedinbase
are removed thisand lumped
iterative models.
process. The
targeted base model is the one achieved when all assumptions that have been made are removed in
this iterative process.
Appendix A discusses this methodology in an example of a system engineering application, an
autonomous unmanned vehicle fleet providing emergency ambulance service.

Figure4.4.Development
Figure Development of lumped
lumpedmodels
modelsvia
viaassumption
assumptionaddition/removal.
addition/removal.

Appendix Capabilities
4. Simulation A discussesand
thisTools
methodology in an example of a system engineering application,
an autonomous unmanned vehicle fleet providing emergency ambulance service.
Many of the enterprise architecture (EA) modeling tools (for example, Sparx Enterprise
Architect, NoMagic MagicDraw/Cameo, IBM Rational/Raphsody) adhere to the Unified Modeling
4. Simulation Capabilities and Tools
Language (UML), Systems Modeling Language (SysML), Business Process Modeling Notation
Many DoDAF
(BPMN), of the enterprise
and UAFarchitecture (EA)
specifications [12].modeling toolshas
Much work (for example,
gone Sparx Enterprise
into developing Architect,
the modeling
NoMagic MagicDraw/Cameo,
infrastructure IBM Rational/Raphsody)
so that the multi-domain knowledge can be adhere
centrallytomanaged
the Unified
andModeling Language
shared across all
the stakeholders. However, the simulation capability required to experience and experiment with the
dynamic behavior of such a model is quite rudimentary as the modeling capability still maps to
traditional systems engineering practices. This further raises questions on the adequacy of widely
understood modeling notations such as UML, SySML and BPMN for developing a complex adaptive
systems model and the simulation infrastructure required for experimentation with such a model [5].
Systems 2018, 6, 40 11 of 18

(UML), Systems Modeling Language (SysML), Business Process Modeling Notation (BPMN), DoDAF
and UAF specifications [12]. Much work has gone into developing the modeling infrastructure so
that the multi-domain knowledge can be centrally managed and shared across all the stakeholders.
However, the simulation capability required to experience and experiment with the dynamic behavior
of such a model is quite rudimentary as the modeling capability still maps to traditional systems
engineering practices. This further raises questions on the adequacy of widely understood modeling
notations such as UML, SySML and BPMN for developing a complex adaptive systems model and the
simulation infrastructure required for experimentation with such a model [5]. Consequently, much of
the work done in computational analysis of system of systems M&S is done in the academic and
research community which leads the development of the accompanying infrastructure employing
High Performance Computing (HPC) and cloud environments. The SES/DEVS methodology has been
employed in a wide range of modeling and simulation applications for a range of complex systems
(known by various other labels such as adaptive systems, system of systems, complex adaptive systems,
networked system of systems, or a combination of these) through simulation and experimentation.

5. Discussion and Way Forward

Two arguments need to be made: MBSE without simulation and MBSE with simulation. Before
making the arguments, we must state unequivocally that modeling and simulation are distinct activities.
Modeling facilitates understanding of phenomena (both natural and artificial) and helps develop an
understanding (both personal and shared). This understanding when coupled with traditional systems
engineering practices gave way to the development of MBSE in its current state. Simulation subsumes
modeling, i.e., simulation is operational only when there exists a model to execute on a platform
(e.g., mental, collaborative, computational). This execution affords experimentation with the model
and provides opportunities to experience the “model” in various settings (for example, Live, Virtual
and Constructive environments [14].
MBSE without simulation, henceforth, involves effort spent in the development of the model.
In SoS, CAS, CASOS, etc. settings, due to a large number of stakeholders, this activity takes on a
whole new meaning where developing a shared understanding is an achievement in itself. IT-enabled
modeling environments commercially available provide the needed centralized repository and model
editing environments to facilitate model development. The prime objective of this activity is to bring the
stakeholders on the same page. In this regard, MBIS can exploit the immersive powers of storytelling
to convey an evolving system design and concept of operations to technically unsophisticated
stakeholders [24].
Between the MBSE without simulation and MBSE with simulation is the realm of executable models.
Formal methods are applied in this model, which lead to software implementation. This enables testing
and verification of systems under investigation during the model runtime. While they are not supported
by experimentation infrastructure, indeed they do allow experience with the system under study.
MBSE with simulation affords experimentation and experience with the model. Simulation
engineering requires an advanced computer science theory, methods and techniques to provide
a computational substrate for the model to execute. When simulation engineering is coupled
with systems theory to develop the computational platform, we get a composable simulation
platform. DEVS is such an example. In SoS, CAS, CASOS, etc. settings, the computational
platform becomes an explicit engineering exercise as new domains are brought in the simulation
environment [15]. The prime objective of this activity is to experiment with the model and gain
experience in understanding model’s behavior. Combining M&S with MBIS can enhance the ability of
models in virtual worlds to foster discovery of previously unknown interactions and dependencies
among system elements and between the system and the environment [24].
Moving forward, as long as one adheres to the primary objectives, both MBSE without simulation
and MBSE with simulation are worthy efforts that employ the art of modeling and the simulation
technology to develop an abstract understanding of the system under consideration and experiment
Systems 2018, 6, 40 12 of 18

with it. Current MBSE practices must extend to incorporate both multiresolution and multi-perspective
modeling within a holistic approach to contribute to complex systems engineering. MBSE, even with
simulation, is inadequate to support complex systems engineering. Complexity Science principles
incorporating concepts like nonlinearity, emergent behavior, network connectivity, etc. is being brought
in to augment MBSE practices with DEVS [17,36] and efforts are underway to develop a comprehensive
methodology for their application to next generation complex systems such as Internet of Things (IoT)
and Cyber Physical Systems (CPS). SES and DEVS provide foundational theory and technology to
engineer M&S-based complex systems in all of their different flavors (SoS, CAS, CASOS, etc.).
We examine these challenges as presentation of a way forward for DEVS M&S in the context of
the roadmap [37] formulated by the International Council on System Engineering (INCOSE).
Along these lines, Zeigler, Mittal and Traoré [31] have identified strong requirements that must be
satisfied to enable DEVS-based M&S to be practiced at its most productive level. Table 2 considers
these developments in support of maturity levels 4 and 5 in Table 3 in the light of the multi-resolution
methodology discussed above.

Table 2. DEVS-based M&S Developments to support maturity levels 4 and 5 of Table 3.

Development Description
The framework provides a multiperspective methodology for developing
coupled models of components from various formalisms capable of
Deal with SoS nature
expressing the different perspectives needed for SoS together with holistic
abstractions that support integration and coordination [17,38–41].
DEVS-based M&S includes the macro level facets that properly organizes
Develop an effective
the CASOS domain and supports refinement into more detailed
organizational ontology
components at the meso and micro levels.
The SES/MB (Model Base) supports hierarchical composition of the
coupled model resulting from pruning that selects from the combinatorial
Enable the ontology to support
family of possible compositions described by the SES. The DEVS
combinatorial model compositions
formalism which can encompass models expressed in various formalisms
and exploration
typically found to be useful in simulation studies [38]. MBIS can enhance
exploration of model behaviors in LVC settings [24].
Include the major facets major facets to The 4-layered ontology highlights at each layer a generic key
ensure representation of all levels (macro, characteristic. It capitalizes on the abstractions used for the simulation of
meso, and micro) of behavior, the entire targeted domain.
An example in healthcare [31] illustrates models spanning health
diffusion, resource allocation, provider and provision modeling,
Include a large spectrum of models for
population diffusion, spatial models including agent-based models at
combinatorial composition
individual and higher level abstractions including
coordination mechanisms.
The MSF includes experimental frames that can specify, collect and
Instrument the complex system to support aggregate the information for higher levels in a multiperspective model.
on-going high quality data The simulation infrastructure guarantees correct execution of the
composed model and the behaviors in a transparent manner [5,17].
A wide variety of mechanisms is available at different levels of abstraction
and computational complexity with typical parallels drawn to biologically
inspired learning and evolutionary processes including activity-based
credit assignment, unsupervised techniques (e.g., clustering, rule mining)
Include pervasive incremental
and reinforcement learning [42–45]. These can be based on the premise
automated learning
that new system states are being continuously captured in timely
snapshots of data and added to an accumulated repository representing
the system knowledge supporting iterative training employing updates in
the system behavior.
Systems 2018, 6, 40 13 of 18

Table 3. Way forward for DEVS M&S in relation to INCOSE Roadmap.

Correlation with DEVS

Maturity Level INCOSE MBSE Roadmap
M&S Development
DEVS model standard and DEVS
1 Emerging MSBE Standards
simulation protocol standard
Experimental Frame representation of
Matured MBSE methods and metrics: Integrated
2 metrics, DEVS models for
System (Hardware/Software) models
Integrated Systems
Architecture Models Integrated with Simulation, DEVS framework for Architecture models
3
Analysis, and Visualization and support for analysis and visualization
4 Defined MBSE theory, ontology and formalisms See Table 2
Distributed and secure model repositories crossing
5 See Table 2
multiple domains

6. Conclusions
As stated before, in the IoT and CPS era, the existing MBSE toolsets lack the needed simulation
analysis and experimentation capabilities leading to epistemological emergent behaviors. Accordingly,
there arises the goal of trying to manage the bad aspects of emergence while preserving its good
qualities. This is reminiscent of Whitehead and Russell’s [46] attempts to control self-reference in
Formal Mathematics in the Principia Mathematica, which they eventually proved to be paradoxical in
nature. Principia’s solution of hierarchical set constructions may suggest a way forward in the SoS case
and the DEVS formalism offers a ready-made solution for hierarchical model construction justified by
closure under coupling [6,16].
MBSE in its current state is very much tied to traditional systems engineering and needs to be
expanded to incorporate complex systems engineering practices. It seems clear that we need to get a
better handle on the whole SoS life-cycle with a more deliberate combined MBSE/DEVS approach.
This will help us focus on the problem and better understand the five attributes of SoS [47] that underlie
and interact to induce emergence in a formally defined system with explicit coupling information [48].
Still, we should recognize enormous obstacles that must be overcome to achieve these visions.
Progress may require new ways of thinking about systems that truly enable them to be developed
with reusable components, eventually leading to composable M&S solutions. We must become able to
identify the limitations in dealing with Big Data and limitations in dealing with its multi-dimensional,
hierarchical, and uncertain nature. Here, we have considered the problems raised by MBSE taken
as a modeling activity without the support of full strength integrated simulation capability and the
potential for, and possible forms of, closer integration between M&S and MBSE as expressed in
augmentation of the INCOSE roadmap for MBSE maturation with requirements for DEVS-based M&S
evolution. Working to put the infrastructure in place to meet these requirements will move both
systems and M&S communities along realistic paths towards realizing the INCOSE roadmap.

Author Contributions: All authors contributed equally to Conceptualization, Methodology, Writing—Original

Draft Preparation, Writing—Review and Editing.
Conflicts of Interest: The authors declare no conflict of interest.
Disclaimer: The author’s affiliation with the MITRE Corporation (McLean, VA, USA) is provided for identification
purposes only and is not intended to convey or imply MITRE’s concurrence with, or support for, the positions,
opinions, or viewpoints expressed by the author. Approved for Public Release, Distribution Unlimited
(Case: PR_18-2996-1).

Appendix A. Multifaceted Model Family Example

As suggested in the main text, the focus of our attention is on CASOS, complex adaptive systems
of systems. Here, we briefly present an example to illustrate the emphasized attributes. Following
Dahmann [3], consider the problem of first responders to catastrophic events providing emergency
rescue and relief. Such a service requires coordination of multiple systems (so is an SoS) with complex
Systems 2018, 6, 40 14 of 18

interconnected networking and requires adaptation to the different challenges posed by different calls
to service. Now, imagine that such a service is to be supported by a fleet of autonomous unmanned
vehicles (UAV) responsible for all needed transportation tasks. We restrict the focus further to the system
engineering design of such an SoS providing emergency ambulance service, as an application of the kind
of multifaceted M&S methodology required to effectively deal with such a complex adaptive SoS (CASOS).
First, consider a short list of categories of objectives relevant to the design and implementation of
a new ambulance system employing unmanned autonomous vehicles (UAVs) as outlined in Table A1.
An all-inclusive model would be able to provide the basis for decision making in each of these
categories. However, the impediments to constructing such a comprehensive model make it a near
impossibility. Instead, we can envisage a collection of partial models, each oriented to one or more
objectives [49].
The following is a subset of models that were developed to address the objectives 1–3,
and 8–10 just listed:

• UAVMotion is a discrete event model representing the motion of the UAVs as agents in space
employing only the kinetic parameters of the vehicles and the random space-time distribution of
requests to get a first order prediction of the number required to meet the demand.
• MarkovDutyCycleCTM represents the duty cycle of a typical UAV as a Markov stochastic process
with a small number of states representing its location as at the depot or in the service area and
able or not to provide service.
• Multiwork represents the UAVs as individual servers in a discrete event model with a simple
bidding protocol to coordinate response to incoming requests with the servers progressing through
a duty cycle consistent with the Markov model.
• Hierarchical Composition elaborates on the Multiwork model by incorporating states of the
vehicles (e.g., carrying patient) that bear upon speed of travel and available fuel. Hierarchical
structure results from representation of UAV as itself a composite with components representing
coordination protocols, kinetics, and fuel consumption.
• Design for Adaptive Sustainment is discussed later.

Table A1. Objectives relevant to system engineering of a UAV-based Emergency Ambulance Service.

# Objectives Models Needed

Kinetic models of UAVs, capacities for carrying
Determine travel and payload requirements
1 medical appliances
for UAVs and personnel
Paramedic capabilities
Select locations of depots and
2 Real estate cost, distances involved, traffic characterization
deployment sites
Centralized vs. distributed decision making of which UAV
3 Response policy optimization
to handle emergency call
4 Marketing Environment: consumer tastes, competition
5 Safety assurance Design of alarms, escape routes, fail-safe plans
Coordination mechanisms, communication
6 Interfacing with existing systems
protocols, interoperability
7 Patient satisfaction Waiting time, comfort, etc.
Types and frequencies of medical emergencies to be treated
8 Determine Emergency response required
Response timelines requirements and how to meet them
Ability to adapt to changing environments associated with
9 Autonomous adaptive behavior
different catastrophic events
System sustainment, maintenance Pricing of services, costs of equipment, investment capital,
10
and evolution long term trends, unused capacity, growth potential
Systems 2018, 6, 40 15 of 18

These models were developed in the order of presentation above following the methodology
in Figure 4. The first model assumes vehicles are essentially point elements moving in space with
abilities to respond to requests without coordination in a neighborhood. The second is a highly lumped
model that represents the fleet in an ensemble sense similar to the ideal gas laws of physics. The third
introduces treating vehicles as individuals requiring coordination to provide service, while the fourth
elaborates on this representation to introduce more of the required constraints.
The fifth model directly relates to modeling of CASOS for system engineering design. Such a
design might search for architectures in a trade-off space involving size and cost. Here, for example,
half the number of UAVs at, say, half the cost, might be enough to ensure a response time that is
only 10% higher than nominally specified. However, rather than the systems engineer having to
determine the fleet size prior to fielding, s/he might design upper/lower brackets within which
to constrain an adaptive plan. This might be similar to the way in which the number of Uber
drivers in a town adjusts to its passenger demand. Such “design for adaptive sustainment” objectives
call for inclusion of models of the environment in which adaptation is occurring as well as of the
mechanism mediating the process. A wide variety of such representations is available at different levels
of abstraction and computational complexity with typical parallels drawn to biologically inspired
learning and evolutionary processes [43–45]. One possibility that seems especially apt here (and is
rarely considered) rests on the analogy to the carrying capacity of an ecosystem for a member species.
Here, a species dynamically adjusts to a population size in which it is in equilibrium with the resources
necessary to sustain it [42]. Analogously to build in such self-adjustment into a fleet of UAVs requires
an economic mechanism in which UAVs must “earn their keep” and means for infusion/withdrawal of
UAVs to/from the operating fleet (a direct illustration of supply/demand in the ontology of Figure 2).
Such adaptive models can be tested in simulation and fielded with little change in code using model
continuity methods.
The assumptions, limitations, and contribution of each model are listed in Table A2.
The developmental progression illustrates that addition of assumptions and removal of constraints
does not necessarily proceed in a monotonic manner but in a manner that is guided by the incremental
needs to establish bounds on predictions to come later and to develop easier structural scaffolds for
later construction.

Table A2. A subset of models developed for objectives 1–3 and 8–10 of Table A1.

Model Assumption Limitation Contribution

Kinetic spatial distribution of No duty cycle
Predicts number of UAVs
requests, policy employs representation No
UAVMotion required under fundamental
visibility and startup No
spatial properties
spatial properties product properties
Above limitations +
Combined stochastic and Predicts numbers required
Lumps UAVS into single
MarkovDutyCycleCTM deterministic representation of under simple duty
number that determines
duty cycle cycle approximation
service rate
UAVs individually
represented as servers in duty No product More refined prediction
Multi-workflow cycle, UAV distance to request properties—capacity and including performance/cost
employed, policy employs abstracted work cycle from Pareto frontier
visibility & spatial properties
No spatial Good solution approximation
All requirements 1–3 and 8
Hierarchical representation, no under service
and constraints accounted for
composition account of requirements/constraints—
to first approximation
energy utilization agrees with earlier predictions
Extends UAVMotion model Same as UAVMotion Can predict dynamics of
Design for Adaptive with economic UAV “earn plus simplification of “carrying capacity” including
Sustainment their keep” mechanism and economics and fleet fleet size equilibrium in fixed
modification of fleet size size modification demand environment
Systems 2018, 6, 40 16 of 18

References
1. Wymore, A.W. Model-Based Systems Engineering; CRC Press: Boca Raton, FL, USA, 1993.
2. Wymore, A.W. A Mathematical Theory of Systems Engineering: The Elements; Krieger: Huntington, NY, USA, 1967.
3. Dahmann, J. Systems Engineering Guide; MITRE Publication: McLean, VA, USA, 2018.
4. Mittal, S. Model Engineering for Cyber Complex Adaptive Systems. In Proceedings of the European
Modeling and Simulation Symposium, Bordeaux, France, 10–12 September 2014.
5. Mittal, S.; Martin, J.L.R. Simulation-based Complex Adaptive Systems. In Guide to Simulation-Based Disciplines:
Advancing Our Computational Future; Mittal, S., Durak, U., Oren, T., Eds.; Springer: New York, NY, USA, 2017.
6. Mittal, S. Emergence in Stigmergic and Complex Adaptive Systems: A formal Discrete Event Systems
perspective. Cogn. Syst. Res. 2013, 21, 22–39. [CrossRef]
7. Mittal, S.; Rainey, L.B. Harnessing Emergence: The Design and Control of Emergent Behavior in System of
Systems Engineering. In Proceedings of the Summer Computer Simulation Conference, Chicago, IL, USA,
24–27 July 2016.
8. Orellana, D.; Madni, A. Extending model based systems engineering for human machine interaction
analysis and fault tolerant design. In Proceedings of the Infotech@Aerospace, Garden Grove, CA, USA,
19–21 June 2012.
9. Tolk, A.; Diallo, S.; Mittal, S. The Challenge of Emergence in Complex Systems Engineering. In Emergent
Behavior in Complex Systems Engineering: A Modeling and Simulation Approach; Mittal, S., Diallo, S., Tolk, A., Eds.;
John Wiley & Sons Ltd: Hoboken, NJ, USA, 2018.
10. Mittal, S.; Diallo, S.; Tolk, A. Emergent Behavior in Complex Systems Engineering: A Modeling and Simulation
Approach; John Wiley & Sons Ltd: Hoboken, NJ, USA, 2018.
11. Mittal, S.; Durak, U.; Oren, T. Guide to Simulation-Based Disciplines: Advancing Our Computational Future;
Springer: New York, NY, USA, 2017.
12. Svyatoslav, K. The Practice of Enterprise Architecture: A Modern Approach to Business and IT Alignment;
SK Publishing: Melbourne, Australia, 2018.
13. Zeigler, B.P.; Nutaro, J.J. Towards a Framework for More Robust Validation and Verification of Simulation
Models for Systems of Systems. JDMS Appl. Methodol. Technol. 2016, 13, 3–16. [CrossRef]
14. Diallo, S.; Mittal, S.; Tolk, A. Research Agenda for Next Generation Complex Systems Engineering.
In Emergent Behavior in Complex Systems Engineering: A Modeling and Simulation Approach; John Wiley & Sons
Ltd: Hoboken, NJ, USA, 2018.
15. Mittal, S.; Zeigler, B.P. Theory and Practice of M&S in Cyber Environments. In The Profession of Modeling and
Simulation: Discipline, Ethics, Education, Vocation, Societies and Economics; Tolk, A., Oren, T., Eds.; John Wiley &
Sons Ltd: Hoboken, NJ, USA, 2017.
16. Zeigler, B.P.; Muzy, A.; Kofman, E. Theory of Modeling and Simulation, 3rd ed.; Academic Press: New York, NY,
USA, 2018.
17. Mittal, S.; Martin, J.L.R. Netcentric System of Systems Engineering with DEVS Unified Process; CRC Press: Boca
Raton, FL, USA, 22 February 2013.
18. Camus, B.; Paris, T.; Vaubourg, J.; Presse, Y.; Bourjot, C.; Ciarletta, L.; Chevrier, V. Co-simulation of
cyber-physical systems using a DEVS wrapping strategy in the MECSYCO middleware. SAGE J. Simul. 2018.
[CrossRef]
19. Zeigler, B.P.; Marvin, J.W.; Cadigan, J.J. Systems Engineering and Simulation: Converging Toward Noble
Causes. In Proceedings of the 2018 Winter Simulation Conference, Gothenburg, Sweden, 9–12 December 2018.
20. Bocciarelli, P.; D’Ambrogio, A.; Giglio, A.; Paglia, E. Model Transformation Services for MSaaS Platforms.
In Proceedings of the Model-driven Approaches for Simulation Engineering Symposium, Baltimore, MD,
USA, 15–18 April 2018.
21. Amissah, M.; Toba, A.L.; Handley, H.A.H.; Seck, M. Towards a Framework for Executable Systems Modeling:
An Executable Systems Modeling Language. In Proceedings of the Model-Driven Approaches for Simulation
Engineering Symposium, Baltimore, MD, USA, 15–18 April 2018.
22. Aliyu, H.O.; Maïga, O.; Traoré, M.K. The high level language for system specification: A model-driven
approach to systems engineering. Int. J. Model. Simul. Sci. Comput. 2016, 7. [CrossRef]
Systems 2018, 6, 40 17 of 18

23. Abdurrahman, A.; Sarjoughian, H. Model-Driven Time-Accurate DEVS-Based Approaches for Design.
In Proceedings of the Model-Driven Approaches for Simulation Engineering Symposium, Baltimore, MD,
USA, 15–18 April 2018.
24. Madni, A.M. Transdisciplinary Systems Engineering: Exploiting Convergence in a Hyper-Connected World;
Springer: Heidelberg, Germany, 2018.
25. Kavak, H.; Padilla, J.; Lynch, C.; Dialio, S. Big Data, Agents, and Machine Learning: Towards A Data-Driven
Agent-Based Modeling Approach. In Proceedings of the Annual Simulation Symposium, Baltimore, MD,
USA, 15–18 April 2018.
26. Wang, F.; Zhang, J.J.; Zheng, X.; Wang, X.; Yuan, Y.; Dai, X.; Zhang, J.; Yang, L. Where Does AlphaGo Go:
From Church-Turing Thesis to AlphaGo Thesis and Beyond. IEEE/CAA J. Autom. Sin. 2018, 3, 113–120.
27. Zeigler, B.P. Multifaceted Modeling and Discrete Event Simulation; Academic Press: Cambridge, MA, USA, 1984.
28. Pawletta, T.; Schmidt, A.; Zeigler, B.P.; Durak, U. Extended Variability Modeling Using System Entity
Structure Ontology Within MATLAB/Simulink. In Proceedings of the 49th Annual Simulation Symposium,
Pasadena, CA, USA, 3–6 April 2016.
29. Choi, S.H. Ranking and Selection Framework for Efficient Reverse Simulation. Ph. D. Dissertation, Korea
Advanced Institute of Science and Technology, Daejeon, Korea, 13 June 2018.
30. Zeigler, B.P. Simulation-based Evaluation of Morphisms for Model Library Organization. In Model
Engineering for Simulation; Zhang, L., Lie, L., Zeigler, B.P., Eds.; Elsevier: New York, NY, USA, 2019.
31. Zeigler, B.P.; Mittal, S.; Traoré, M.K. Fundamental Requirements and DEVS Approach for Modeling and
Simulation of Complex Adaptive System of Systems: Healthcare Reform. In Proceedings of the Symposium
on Modeling and Simulation of Complexity in Intelligent, Adaptive and Autonomous Systems, Baltimore,
MD, USA, 15–18 April 2018.
32. Jepperson, R.; Meyer, J.W. Multiple Levels of Analysis and the Limitations of Methodological Individualisms.
Sociol. Theory 2011, 29, 54–73. [CrossRef]
33. Davis, P.K.; Bigelow, J.H. Experiments in Multi Resolution Modeling (MRM); Rand: Santa Monica, CA, USA, 1998.
34. NPS Faculty. Aggregated Combat Models. Available online: http://faculty.nps.edu/awashburn/
Washburnpu/aggregated.pdf (accessed on 3 April 2017).
35. Davis, P.K.; Hillestad, R. Families of Models that Cross Levels of Resolution: Issues for Design, Calibration
and Management. In Proceedings of the Winter Simulation Conference, Los Angeles, CA, USA,
12–15 December 1993.
36. Martin, J.L.R.; Mittal, S. Model Management and Execution in DEVS Unified Process. In Model Engineering
for Simulation; Zhang, L., Zeigler, B.P., Laili, Y., Eds.; Academic Press: Cambridge, MA, USA, 1 March 2019.
37. MBSE Wiki. Available online: http://omgwiki.org/MBSE/doku.php (accessed on 11 November 2018).
38. Zeigler, B.P.; Sarjoughian, H. Guide to Modeling and Simulation of System of Systems; Springer: New York, NY,
USA, 2017.
39. Zeigler, B.P.; Nutaro, J.J.; Seo, C. Combining DEVS and model-checking: concepts and tools for integrating
simulation and analysis. Int. J. Simul. Process. Model. 2017, 12. [CrossRef]
40. Gholami, S. Combined DEVS multiresolution simulation and model checking. In Proceedings of the Winter
Simulation Conference, Las Vegas, NV, USA, 3–6 December 2017.
41. Neto, V.V.G.; Manzano, W.; Kassab, M.; Nakagawa, E.Y. Model-Based Engineering & Simulation of
Software-Intensive Systems-of-Systems. In Proceedings of the 12th European Conference on Software
Architecture: Companion Proceedings, Madrid, Spain, 24–28 September 2018.
42. Arrow, K.; Bolin, B.; Costanza, R.; Dasgupta, P.; Folke, C.; Holling, C.S.; Jansson, B.O.; Levin, S.; Mäler, K.G.;
Perrings, C.; Pimentel, D. Economic Growth, Carrying Capacity, and the Environment Science. Ecol. Econ.
1995, 15, 91–95. [CrossRef]
43. Sutton, R.S.; Barto, A.G. Reinforcement Learning: An Introduction, Bradford Book; MIT Press: Cambridge, MA,
USA, 1998.
44. Muzy, A.; Zeigler, B.P. Activity-Based Credit Assignment Heuristic for Simulation-Based Stochastic Search in
a Hierarchical Model Base of Systems. IEEE Syst. J. 2017, 11, 1916–1927. [CrossRef]
45. Vasbinder, J.W.; Gao, H. Selected Papers of John H. Holland: A Pioneer In Complexity Science; World Scientific
Press: Singapore, 2018. [CrossRef]
46. Whitehead, A.N.; Russell, B. Principia Mathematica; Cambridge University Press: Cambridge, MA, USA, 1913.
Systems 2018, 6, 40 18 of 18

47. Maier, M.W. Architecting Principles for Systems-of-Systems. J. Int. Counc. Syst. Eng. 1998, 1, 267–284.
[CrossRef]
48. Ören, T.; Mittal, S.; Durak, U. Induced Emergence in Computational Social Systems Engineering:
Multimodels and Dynamic Couplings as Methodological Basis. In Emergent Behavior in Complex Systems
Engineering: A Modeling and Simulation Approach; Mittal, S., Diallo, S., Tolk, A., Eds.; John Wiley & Sons Ltd:
Hoboken, NJ, USA, 2018.
49. Zeigler, B.P. Structuring the Organization of Partial Models. Int. J. Gen. Syst. 1978, 4, 81–88. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).