You are on page 1of 9

An Agent-Based Modeling Approach to Creating More

Resilient Littoral Combat Architectures


James Behmer, Kolawole Ogunsina, Parth Shah, Suhas Srinivasan
School of Aeronautics and Astronautics, Purdue University
West Lafayette, IN, USA
Email: jbehmer@purdue.edu, kogunsin@purdue.edu, shah66@purdue.edu, suhas@purdue.edu

Abstract—Littoral combat is a critical component of naval war- independent and useful systems are integrated into a larger
fare that has gained importance since the late 20th century due system that delivers unique capabilities.” Additionally, the
to a shift from matched blue-water combat to agile, asymmetric following characteristic traits were proposed by Maier [3] to
warfare near the coast. With the increased relevance of this aid in identification of what might distinguish an SoS from a
type of naval combat, the ability of a networked naval warfare monolithic system:
System-of-Systems (SoS) to operate effectively given the chal-
lenges of operations in the littoral zone becomes imperative.
Resilience in an SoS in the face of a variety of threats is vital. Operational Independence Constituent systems may have
These threats may take various forms: traditional warfare, their own tasks to fulfill outside of the context of the SoS.
cyber attacks, or communications breakdowns. The challenge Managerial Independence Units may have a unique tasking
for any SoS practitioner is identifying such architectures from a by their respective operators/owners.
vast design space. We define an architecture as the constituent
systems in the SoS and the connections among these systems. Some heuristics were also proposed in [3] on the effective ar-
We seek to apply an agent-based modeling approach to create an chitecture of SoS namely the concept of stable intermediate
architecture evaluation tool that leverages the modular nature of
an SoS to quickly conduct trade studies and analyses to identify forms– the SoS must be able to perform its functions even
architectures that are resilient to a variety of threats. The if not at its complete/optimal configuration, policy triage–
simulation tool is built on the principles of the Open Systems the designers need to exact just the right amount of control
Architecture put forth by the Department of Defense (DoD). We over the model design, leveraging at the interfaces– with
present an analysis that seeks to find a suitable SoS architecture the components having their own independence the design of
configuration that maintains operability and performance dur- the SoS boils down to how they communicate with each other,
ing the loss of command authorities. This loss can be attributed and finally ensuring cooperation among the constituent sys-
to battle casualties or other forms of threats including commu- tems. Keeping these in mind, we aim to look at the premise
nications failures. We demonstrate that distributing managerial of littoral combat from an SoS perspective.
control in the architecture improves, to some extent, SoS perfor-
mance in terms of time to threat elimination. This single analysis
demonstrates how SoS practitioners can use agent-based models Context and Motivation
to quickly construct and evaluate potential architectures against The term ‘Littoral Combat’ refers to naval engagements close
a variety of threats and disruptions, ultimately building resilient to the shoreline generally in support of or related to amphibi-
SoS architectures that perform to prescribed levels even in the
face of network failures and attacks. ous operations. While by no means a new aspect of naval
warfare2 it has gained increased relevance in the late 20th and
21st century. This could possibly be due to a shift from evenly
matched blue-water engagements to a more asymmetric form
TABLE OF C ONTENTS against smaller or non-state actors. This could make an
already challenging scenario all the more difficult especially
1. I NTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 when one considers the possibility of civilian traffic in the
2. B UILDING THE M ODEL . . . . . . . . . . . . . . . . . . . . . . . . . . 2 same waters.
3. A NALYSIS - A RCHITECTURE P ERFORMANCE . . . 5
4. R ESULTS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 To realize this increased relevance, the US Navy has com-
5. C ONCLUSION , S UMMARY AND F UTURE D E - missioned the Freedom[5] and Independence[6] classes of
VELOPMENT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Littoral Combat ships (LCS). Along with the ’Sea Lance
System’ [7] the LCS is expected to feature prominently in
the Navy’s future engagements. In this context, our proposed
1. I NTRODUCTION objective is to develop a model and simulation for potential
littoral engagements that implements both directed and ac-
Designing a suitable System-of-Systems (SoS) architecture knowledged forms of managerial control. The acknowledged
poses strong challenges owing to the need to balance the variant of the SoS would require that subsystems maintain
independent managerial control of constituent elements as their own management akin to the SoS. This should ensure
well as their operational interdependence. To gain a better cooperation among subsystems as each is now incentivized
idea of what makes SoS architecture design challenging, it to perform more effectively.
may be necessary to understand what constitutes an SoS and
how it differs from a system. A system is defined by Gibson
et al. in [1] as “a set of elements so interconnected so as to aid Our motivation for this project is the belief that a major
in driving toward a defined goal.” In comparison the Defense challenge for a littoral combat naval warfare SoS is the
Acquisition Handbook (DAG) (2008)[2] seeks to describe an
SoS as “a set or arrangement of systems that results when 2 Notable examples include the Gallipoli landings(1915), the invasion of
Normandy (1944), and the invasion and counter-invasion of the Falkland
978-1-4673-7676-1/16/$31.00 2016
c IEEE Islands (1982)[4].

1
Table 1. ROPE Table for the proposed SoS 2. B UILDING THE M ODEL
Agent-Based Modeling
Levels Resources Operations Policy Economics
α Sensors, Commun- Communi- Unit Systems-of-systems and large-scale complex systems such as
Con- icate with cations Operat- the stock market are difficult to model analytically due to the
trollers, other Rules ing and complex nature of interactions among constituent systems.
Com- agents to Training Additionally, the observed behavior in these systems adheres
manders, generate Cost to the notion, “the whole is greater than the sum of its
Weapons a threat parts”. Agent-based modeling (ABM) strives to capture
map the complex behaviors using a bottom-up approach where
individual systems in the SoS are expressed as agents in the
β Platforms: Maneu- Rules of Mainte- model. This approach provides a more natural description
LCS, vering engage- nance of the system, is flexible, and captures emergent phenomena
Aircraft, and ment and that an SoS practitioner would be most interested in[9]. An
Sub- engaging operating agent model receives information from the environment and
marines targets costs acts using a set of internal rules and principles. These basic
principles are replicated for all agents in a system. Axtell
γ Theater Deploy- Strategic Defense identifies three potential cases where agent-based modeling
Com- ment Policy, Budget can be beneficial[10]:
mand deci- Order of
Authority sions, site Battle 1. The model can be expressed using analytical means. In
planning this case, the agent-based model can serve to verify the results
obtained from analytical means. This however is not the most
effective use of agent-based modeling and is rarely employed.
2. The model can only partially be expressed analytically.
In this case, a numerical agent-based model can help comple-
ment the analytical expression of the system to obtain a result
decentralization of Command & Control (C2 ). If all control that is otherwise unattainable.
of the SoS is delegated to one unit (for example a single 3. The model is intractable. This case is especially true in
LCS), the risk of complete system failure would be very large-scale complex systems. Using an agent-based model
high if that particular system is removed from the scenario. can give insight into key behaviors and a better understanding
Distributing control among similar C2 capable systems might of the system as whole.
help mitigate this risk. This shift in control is also in line
with the US Navy’s initiative to move towards a net-centric Case #3 is directly applicable to SoS as it is difficult to capture
warfare (NCW) model as described by its four tenets [8]: the independent behaviors of each individual system and the
interactions among them. Additionally, many different types
of agents can easily be constructed by changing internal
1. Robust networking improves information sharing parameters that govern how they make decisions. This makes
2. Collaboration and information sharing improves the qual- ABM particularly well-suited for littoral combat applications
ity of data and shared situational awareness where an SoS practitioner as able to perturb specific agent
3. Shared situational awareness enables self-synchronization parameters or methods from a baseline and observe the im-
4. Which leads to an increase in mission effectiveness pact on overall performance. The advantages of ABM in this
domain do not render other methods (e.g. system dynamics,
model-based systems engineering) ineffective. In the realm of
Table 1 shows our layout of the hierarchy of constituent
systems (Resources), their functions within the context of
the SoS (Operations), their controlling rules (Policy) and the
Economics of operation.

In this project, we look at networking multiple β level plat-


forms to produce architectures with different properties.The
architecture design spectrum adopted from [3] is divided as
follows

Directed Constituent systems participate solely to fulfill SoS


objective.
Acknowledged Constituent systems help accomplish SoS
goals, but systems maintain independent objectives as well.
Collaborative Constituent systems contribute voluntarily to Figure 1. An example BMDS scenario illustrating potential
achieve SoS objective platforms
Virtual No central authority or clear SoS objective and all
systems are not always known
defense, a few distinct platforms exist that can be represented
Our main focus is the transition from a Directed to a more by discrete agents in a simulation. Per the Department
Acknowledged architecture and a comparison between them. of Defense Architecture Framework (DoDAF) definition, a
The setup is explained in Section 2 and the performance platform is “a physical structure that hosts systems or system
comparison in Section 4 hardware or software items”[11]. The need to distinguish
2
between a platform and agent will become evident in the
following section. Consider the notional defense scenario
shown in Figure 1. An enemy threat is launched and detected
by ground and space based sensors. Once all tracking require-
ments are met, an interceptor can be launched to intercept
the threat at a future time. This scenario can be represented
by four distinct platforms containing several agents that
perform critical functions that allow a platform to function
as desired. The combination of these platforms and the
interactions among onboard agents construes an architecture.
For example, at minimum, the space-based radar platform
must contain a sensor (agent) that gathers measurements from
its surroundings. Additional onboard agents can leverage the
measurements to estimate a track for a detected threat or the
raw measurements can be relayed to other platforms. Often
times, space-based assets can also serve as communications
relays to other platforms. The following section discusses the Figure 2. Potential DAF Architecture
Discrete Agent Framework, an ABM modeling framework
used for TRITON.

DAF agents are connected to each other. A link allows agents


Numerous agent-based modeling tools exist to simulate large- to communicate and limits the interaction solely to those
scale systems. Net- Logo, Swarm, MASON, and Repast are agents that are connected to one another and performance
just a few examples of ABM tools available to any designer. hinges on how the architecture is built. Figure 2 illustrates
These tools employ a framework and library paradigm [12] a potential architecture that can be created using DAF. Three
that provide a framework to design the agent-based model distinct platforms have agents that serve distinct functions.
and a library of tools to implement the design. Discrete Agent The arrows indicate links among agents governing how in-
Framework (DAF) is an agent-based modeling tool developed formation flows in the architecture. Additionally, the ar-
and maintained at Purdue University’s System-of-Systems chitecture can easily be modified to include new agents or
Laboratory [13]. DAF is built on the Open Systems Architec- even redistributed to change information flow. One of the
ture (OSA) approach, which emphasizes code flexibility and most attractive features of DAF is the ability to evaluate
openness [14] and possesses many unique characteristics that many different architecture options quickly which makes it
make it useful in modeling SoS problems. Built in object- a tractable modeling framework for our application.
oriented MATLAB, DAF provides a generic class of methods
TRITON
and functions that can support domain-specific applications3 .
DAF can be broken into four main components: agents, TRITON was created as a tool that uses DAF’s existing
platforms, messages, and architecture structure. By default, framework for agent based modeling. Our model is devel-
only one agent exists in DAF as it serves as the foundation for oped to look at changing communication architectures and
domain- specific applications to allow developers to create DAF provides a suitable method for that by having reac-
their own agents. Agents are initialized as new classes that tive agents activate upon receipt of messages. In essence,
inherit base properties and methods from the DAF base agent the content of these messages between agents, their routing
class. Other agents can access some of these properties and comprises our idea and implementation of communication
methods while others are private. An agent in DAF contains architecture. The model was designed in a manner conform-
methods and properties that perform a specific function on a ing to the strategy outlined in Table 1. The base agents
platform. Platforms in DAF are a physical instantiation of a (sensors, weapons, controllers) at the α level are grouped
group of agents. DAF is capable of executing the methods into different classes of β level platforms (Submarine, At-
on all agents and platforms at discrete time intervals, but can tack/Recon aircraft and LCS) referred to as the ‘units’ in the
also trigger execution based on messages. Messages in DAF simulations. It should be noted that while there are three main
dictate how information flows among agents. DAF supports classes of platforms, they all bear base properties that can be
two types of messages: broadcast and directed. Much like customized a priori to represent either an air, sea or sub sea
agents, no specific DAF messages exist, but developers are unit. Additionally, many of our candidate metrics– outlined
able to leverage DAF to create messages that are domain- in Table 3 in Section 3 use the time to major events within the
specific. Once a message structure is specified, agents can simulation to characterize SoS performance.
send and receive messages, effectively communicating as
they would in an SoS application. Additionally, agents can
be restricted to receive only certain types of messages and ABM uses different types of agents, which make up different
latency models help simulate the real-world delay associated types of platforms, in an attempt to identify patterns in their
with information transfer between two agents. Agents in behavior. Our model uses the following types of agents:
specific applications are often reactive in nature and thus
only execute when input is provided. This behavior can be Commanders : process threat information; send orders; carry
mimicked in DAF by dictating that agents execute only when out orders
messages are received. Controllers : relay threat information to commanders; carry
out orders
Finally, an architecture in DAF is formed by defining which Sensors : identify threats and send information to controllers
or commanders
3 An in-depth analysis of DAF and its structure falls out of the scope of
this work. For a more detailed analysis, please refer to Fry, Campbell, Each platform in our model consists of a sensor and either a
DeLaurentis, 2015 [14] commander or a controller. Our platforms and the motivation
3
Table 2. Generic Platform Properties around the sensor’s range, the detection capability (pDetect)
is modeled as a Gaussian surface centered at the platform’s
Platform Type Properties location at each instance.
Speed :30kts  
−(distance2 )
pDetect = exp
Submarine pkill :0.9 2 × Sensor Range
Model Basis :Los Angeles
Class This too is compared against a randomly generated threshold
Armament :ASM, Torpedoes to account for unmodeled dynamics that might affect the
Speed :50kts sensor’s detection of a threat. Additionally the following
Littoral Com- pkill :0.2 assumptions are made:
bat Ships Model Basis :Independence
Class • Red forces move only when provoked
Armament :ASM, SAM, Tor- • A platform is disabled and removed from the simulation if
pedoes hit by a weapon
Speed :500kts • Platforms hold constant altitude when maneuvering
Aircraft pkill :0.44 • All messages are relayed at equal cost
Model Basis :F/A-18 E/F • Messages between platforms never fail to send
Armament :ASM,AAM • Sensors have full field of view

Networking, Architectures
The network architectures used in the analysis of the littoral
for their design are listed in Table 2. combat scenario of the Falklands War [17] are discussed in
this section. According to Maier, operational and managerial
independence are vital in establishing the uniqueness of an
In addition to these, two single-role platforms are defined, SoS network architecture [3]. In this vein, three network
namely the anti-ship mines and mine sweeping platform.
The mine sweeping platform is based on the MH-53E Sea architectures namely Directed, Acknowledged and Mixed
Dragon platform [15] and the anti-ship mines are based on the architectures are adopted, based on their operational and man-
Stonefish influence mine [16]. In this scenario, the mines are agerial form of control, in modeling the littoral combat sce-
deployed by the red force and the MH-53E mine sweeper by nario. Currently, most integrated defense systems, especially
the blue force. This arrangement allows for more metrics to those involving naval warfare, utilize a strictly directed SoS
measure effectiveness – the percentage of simulations where architecture as their main form of managerial control. The
all mines are destroyed (Figure 8 and Equation 1) and the component systems are designed and centrally regulated such
average destruction per simulation (Figure 9 and Equation 2). that they carry out specific tasks geared towards the overall
Further details are described in Section 4 goal of the SoS [3]. However, a directed architecture runs a
high risk of being overly controlling, and this may dissuade
existing or potential subsystems from effectively fulfilling
Because agent interfaces define the baseline of the model, their roles in the SoS. An acknowledged SoS architecture
comprises of constituent systems that help to accomplish the
agents can quickly be added or removed as either comman- SoS goal, but the systems maintain their own management
ders or controllers. Furthermore, agents can easily be re- and authority akin to the SoS. This ensures cooperation
wired to represent different networks. This modularity is amongst subsystems, as each subsystem is incentivized to
critical during the analysis phase for it allows us to test perform efficiently with the knowledge that its successful
multiple network configurations. operations ultimately benefit the SoS. The mixed architecture
combines the properties of both directed and acknowledged
Modeling Assumptions architectures [3].
In order to build a suitable model that captures desirable
dynamics of the SoS while being modular and computation-
ally simple, some assumptions are made. Although these We focus our efforts on defensive littoral combat site plan-
assumptions would not hold in a real-life scenario we believe ning, which includes two separate force (red and blue)
them to be reasonable enough to make modeling possible compositions. The attacking red force maintains a constant
without ruining the integrity of our results. network configuration in the analysis as shown in Figure 3,
Weapon dynamics are modeled probabilistically using range, with 2 commanders and 10 controllers representing a strictly
weapon type, and the target platform’s expected defen- directed SoS network architecture. Two of the four littoral
sive/evasive capability. Each platform is capable of carrying combat ships (LCS) are command and control (C2 ) agents,
a mix of one or more of Anti-Ship Missile (ASM), Torpedo, with each C2 agent having a connection with three of the
Surface-to-Air Missile (SAM), and Air-to-Air Missile types. six attack aircraft (AA), one LCS, and both submarines in
Depending on the currently targeted threat, the appropriate the SoS network. The red force network loses half of its
platform is tasked to engage it with the suitable weapon effectiveness if one of the C2 agents is destroyed. Hence, the
type. The effectiveness of the engagement is modeled as a C2 s in the network exhibit high levels of situational aware-
Total Kill Probability (tpk). Clearly tpk would be smaller ness to maintain adequate red force effectiveness. While
for a target further away. This is compounded by the the blue force composition remains constant, the links and
weapon accuracy and defensive capabilities of the target. number of C2 agents within each platform change to allow for
The calculated value of tpk is compared against a randomly the network architecture evolution. Figures 4,5,6 represent
generated threshold value (0, 1] to account for the chance that the directed, acknowledged, and mixed network architectures
an engagement may succeed in difficult conditions as well as for the defending blue force respectively. The directed blue
it failing under very good conditions. force network consists of two LCS, three attack aircraft, one
The sensor effectiveness is similarly modeled. Given a ball reconnaissance aircraft (RA), and one submarine. However,
4
Figure 3. Red Force Architecture Figure 5. Acknowledged Blue Force Structure

Figure 4. Directed Blue Force Structure Figure 6. Mixed Blue Force Structure

objective is to design an SoS for littoral combat site planning,


one of the two LCS is a C2 agent directly connected to the we decided to use reachability to define our metrics. Three
remaining assets in the network. The C2 (LCS) agent has metrics are used in the model:
complete operational control of the SoS, and its elimination
results in zero network effectiveness. The acknowledged
blue force network composition is similar to the directed one, 1. Time to Threat Elimination
except that both LCS, the submarine, and the reconnaissance 2. Number of Red Force (attacking force) platforms remain-
aircraft are all C2 agents. This reduces the susceptibility of ing
the SoS network to point failures as operational control is 3. Number of Blue Force (defending force) platforms re-
distributed among all four C2 agents. Finally, the mixed blue maining
force architecture and acknowledged blue force architectures
are compositionally similar and contain the same number of As stated earlier in Section 1 our goal for the development
C2 agents. However, the C2 agents in the mixed architecture of TRITON is to assess the effectiveness of shifting from a
are less connected to other assets in the SoS network, result- directed architecture to a more acknowledged state on the
ing in a lower level of situational awareness for the command combat resilience of the ‘blue’ force. We’ve mentioned
and control agents. previously in Section 2 that the variable being controlled
for the analysis is the set of interfaces between platforms
for the blue force, replicating the shift from directed to
3. A NALYSIS - A RCHITECTURE
P ERFORMANCE Table 3. SoS Metrics
Approach and Analysis Set-up
Before choosing a modeling approach it is important to Meta-Metric Area SoS Metric
develop a pool of metrics to be measured by our computer Reachability Average time to threat
elimination; time to first
model. These metrics are used to form a rudimentary hy- threat detection
pothesis concerning the behavior of the SoS. The most low- Robustness Time to recognize failure,
level, bare bones metric area is reachability. Reachability where failure is defined as
determines our SoS ability to identify the best performance a percentage of assets elim-
configuration with the given constraints: rules of engage- inated; Recover from nodal
ment, bandwidth, firepower, and mobility. The next area con- failure (resiliency)
cerns uncertainty and is called robustness. The main source Scalability Ratio of time to threat elim-
of uncertainty is nodal failure. If a subordinate system fails ination vs cost of oper-
(i.e. destroyed by enemies), a robust SoS will quickly identify ation; Probabilistic threat
this fact and allocate resources accordingly. In regards to detection based on thresh-
efficiency, scalability can help maintain costs at a level just old (conservatism)
high enough to deal with threats. This is an important area Versatility Ability to adapt to differ-
in minimizing cost. Our SoS must also be applicable to ent types of threats; Change
the general coastline. Therefore, versatility is critical not in threat elimination time
necessarily within a specific conflict, but throughout various over multiple coastal battle
conflicts at different geographical locations. Table 3 summa- spaces
rizes candidates for littoral warfare SoS metrics. Since our
5
acknowledged architectures. This boils down to deciding
which platforms have C2 authority and which platforms the 
C2 platforms command. Sims. All mines destroyed
Complete Destruction = (1)
Each run of the tool covered 2000 seconds of simulation N Simulations
time in approximately 25 minutes. This experiment was
repeated 150 using Monte-Carlo methods totaling to 48 hours Somewhat counter intuitive is the behavior seen where the
of computing time. Although the model performs well for mixed SoS combat architecture destroys all 40 mines more
our purposes, there is room for significant improvement with often than the acknowledged architecture. Figure 9 analyzes
regards to the computational cost. One thing that becomes this result one level deeper by examining the average de-
apparent is that the C2 units in more acknowledged struc- struction across all simulations. For each architecture type,
tures can get by with lower situational awareness owing to 50 Monte Carlo simulations where used to average out the
other similarly capable units operating semi-independently probabilistic calculations in the simulation (Equation 2).
and directing their subordinate controllers (and platforms) to
multiple threats. N
i=1Mines destroyed
Average Destruction = (2)
Asset Placement N Simulations
Unlike Figure 8, in this decomposition, we see the behavior
adheres to the initial suspected behavior that average de-
struction increases when moving to a more acknowledged
state. Correlating back to the meta-metric areas discussed
in the prior sections, this behavior speaks to the robustness
of the SoS architecture. By including more C2 authorities,
more mines are destroyed on average, but at the cost of
reduced situational awareness. This can become costly when
considering the dynamic nature of littoral combat. Consider
a situation where two commanding authorities hold engage-
ment orders for a singular aircraft platform. Conflicting
engagement orders can potentially reduce the effectiveness
of the platform. If the order from Commander A asks the
aircraft to move away from a potential target with engage-
Figure 7. Asset Placement in the Scenario Zone ment probabilities at time t1 and Commander B asks the
target to return shortly after, a split second deviation can
potentially nullify the more promising engagement. The
analysis team suspected such a case when considering why
The location chosen for our scenario zone was the Strait of the mixed architecture displayed more complete destruction
Juan de Fuca that straddles the border between the USA and of mines compared to the acknowledged architecture.
Canada on the west coast. Our reasons for choosing this
are twofold. The first reason being that it provides a good
representation of potential real world operating locations.
The strait appears to be a route of movement for assets
stationed at the Bremerton and Kitsap-Bangor naval bases.
Secondly, the operating zone seems to be fairly uncluttered
in landmasses in and around the area of movement of the
platforms. This is important to us as TRITON does not yet
include capabilities for the platforms to detect and maneuver
against landmasses. Figure 7 shows an example of how the
forces will be laid out. Figure 7 also shows the architecture
types of the red and blue forces for the directed architecture
simulations.

4. R ESULTS
This investigation strives to assess the effectiveness of dif-
ferent blue force architecture types (directed, mixed, and Figure 8. Percent of Simulations where all mines are
acknowledged) against a static red force composition. Each destroyed
simulation contained 40 mines (modeled after Stonefish
mines). In addition to the three metrics (Time to threat
elimination, red and blue forces remaining), the number of
mines destroyed in each simulation serves as an analog to These two figures offered a glimpse into the static effec-
assessing architecture effectiveness. With added command tiveness of the architecture at the end of the simulation,
authorities (C2 capable platforms) the MH-53E helicopter can but does not speak to the effectiveness of the architecture
receive additional tasking commands to remove mines. This over the duration of the simulation. Figure 10 displays the
notion suggests that added situational awareness (in more blue force, red force, and mine fraction as a function of
mixed and acknowledged architectures) would maximize the both time and architecture type. The directed architecture
number of mines destroyed in the simulation. Figure 8 structure is represented using circle markers, mixed with
shows the percentage of simulations where all 40 mines are square markers, and acknowledged with x markers. Several
eliminated as captured in Equation 1. interesting behaviors are seen from this decomposition of the
6
littoral combat architectures using an agent-based modeling
approach. The authors believe that the results shown illustrate
the capability of TRITON in its infancy and also reveal many
avenues for exploration and development. The modular,
adaptable nature of the software tool offers SoS practitioners
a top-level modeling tool to analyze the effectiveness of po-
tential architectures quickly. Potentially viable architectures
identified using TRITON can then be evaluated using higher
fidelity simulations. Often, a goal of agent-based modeling is
to identify emergent behavior from the simulation. Emergent
behavior is seen only when analyzing the interactions among
agents rather than pairing the agents together themselves.
The authors find it enticing to claim emergence regarding the
behavior seen in the analysis, but it is premature given the ma-
turity of the software tool TRITON. Moving forward, several
potential avenues have been identified for development:

Figure 9. Average Destruction per Simulation

data. There is increased blue force survival when transition-


ing to a more Acknowledged SoS architecture, potentially Sensor FOV/FOR + Intelligent Tasking Algorithms
due to the presence of more command capable units in the Currently all sensors are omni-directional and capable of
architecture overall. Additionally, the increased blue force detecting enemy platforms provided they fall in the specified
survival is complemented by a decreased red force survival sensor range. By imposing field-of-view (FOV) and field-of-
or increased red force destruction. The mixed and acknowl- regard (FOR) limits in conjunction with intelligent tasking
edged architectures perform comparably as approximately algorithms, the authors hope to see the effects of slewing and
60% of all red force platforms are destroyed by the end of other congestive effects as multiple threats pass with a single
the simulation as compared to 30% destruction in a directed sensor’s FOV/FOR.
SoS architecture. As noted from Figures 8 and 9, there is Ballistic/Seeking Missile trajectories
increased mine force destruction when transitioning to a more In it’s current state, engagements fly a straight line course to
Acknowledged configuration. an intended target utilizing it’s current position, rather than
projecting a future location. The authors hope to imple-
ment seeking trajectories to reflect the capabilities of existing
fielded missiles.
Intelligent avoidance maneuvering
All platforms, regardless of terrain or obstacles adopt a short-
est distance approach to target. This behavior is not reflective
when considering that submarines may pass through mine
fields and other obstacles in littoral waters. By implement-
ing intelligent movement algorithms, the authors hope that
simulate more robust engagements.
Engagement Plan Updates
Related to seeking trajectories, engagement plan updates
allow platforms to reassess the scenario prior to executing
engagements. Often, due to the dynamic nature of the battle
space, platforms will no longer be engageable based on the
initial estimate. By updating the engagement plan at multiple
time steps, faulty engagements can be eliminated from the
simulation.

Figure 10. Force Composition over Simulation Time

This figure supports the initial notion that shifting to a more


Acknowledged SoS architecture will yield a decrease in the
average time to threat elimination as well as increased blue ACKNOWLEDGMENTS
force survival.
The authors would like to thank their professor, Daniel De-
Laurentis for his unwavering support and guidance. TRITON
5. C ONCLUSION , S UMMARY AND F UTURE was initially built as a final course project for AAE 560 -
System-of-Systems Modeling and Analysis taught at Purdue
D EVELOPMENT University. Using the concepts and modeling principles
Hearkening back to the original goals presented in this taught in the course, TRITON evolved to its current state and
paper, TRITON strives to model and build more resilient will continue to evolve.
7
B IOGRAPHY Suhas Srinivasan completed his Bach-
elor of Engineering in Mechanical En-
gineering at NIT-Trichy, India in 2014.
He is currently pursuing an MS in Aero-
nautics & Astronautics at Purdue Uni-
versity, Indiana specializing in Dynam-
James Behmer is a Member of the Tech- ics and Control and Systems Engineer-
nical Staff at the Aerospace Corporation ing, focusing on nonlinear control tech-
in El Segundo, CA. Jim recently com- niques and design optimization. His pri-
pleted an internship at the Commercial mary area of interest is in the guidance
Spaceflight Federation in Washington, and control of Unmanned Aerial Vehicles and is currently
D.C. where he wrote space policy briefs, working on GPS-denied navigation and localization of Un-
supported meetings on Capitol Hill and manned Systems.
with various government agencies, and
assisted with the advocacy of commer-
cial spaceflight. Prior to CSF, Behmer
spent four months at SpaceX supporting Falcon 9 second R EFERENCES
stage engine tests. He also has flight test and systems en-
gineering experience through internships with Boeing Com- [1] J. E. Gibson, W. T. Scherer, and W. F. Gibson, How
mercial Airplanes in Seattle and Rockwell Collins in Cedar to Do Systems Analysis. Hoboken, NJ, USA: John
Rapids, IA. Jim holds Bachelor’s and Master’s degrees in Wiley & Sons, Inc., may 2007. [Online]. Available:
Aeronautics and Astronautics from Purdue University. http://doi.wiley.com/10.1002/9780470130599
[2] O. A and T. Sse, Systems Engineering Guide for
Systems of Systems, 2008, vol. 36, no. August. [Online].
Available: http://www.acq.osd.mil/se/docs/SE-Guide-
for-SoS.pdf
[3] M. W. Maier, “Architecting Principles for Systems-of-
Kolawole Ogunsina was admitted to Systems,” Systems Engineering, vol. 1, pp. 267–284,
Obafemi Awolowo University, Nigeria 1998.
to study Electrical and Electronic En- [4] D. Wheeland, “Synchronization of Littoral Operations,”
gineering at age fifteen. Driven by his Naval War College, Newport, RI, Tech. Rep. February,
passion for innovation in aviation, he 1994.
transferred to Embry Riddle Aeronau- [5] R. Burke, “LCS Freedom Commissioned,” 2008.
tical University, Daytona Beach FL to [Online]. Available: http://www.navy.mil/submit/
study Aerospace Engineering and grad- display.asp?story{ }id=40822
uated with a Bachelor of Science de- [6] Surface Forces Public Affairs, “USS Independence
gree in May 2014. Kolawole worked as Commissioned,” 2010. [Online]. Available: www.navy.
an intern at Gulfstream Aerospace Corporation and South- mil
west Airlines during his undergraduate education. At the [7] C. Calvano, D. Byers, R. Harney, F. Papoulias, J. Ciezki,
Southwest Airlines Flight Operations Engineering (FLOE) L. T. H. Markle, L. T. R. Trevisan, L. T. T. Barney,
department, he developed One Engine Inoperative proce- L. T. K. Eimers, L. G. Farman, L. T. C. Nash, L. A. Al-
dures that increased the Maximum Takeoff Weight (MTOW) tekin, and L. T. R. Kompatzki, ““SEA LANCE” Littoral
capacity of the Boeing 737 fleet, which translated to airlifting Warfare Small Combatant System,” Naval Postgraduate
three to twenty six additional passengers. In recognition of School, Tech. Rep. January 2001, 2001.
his exceptional support to the flight operations engineering [8] D. Alberts, Information Age Transformation: Getting
department, he was presented with the Southwest Airlines to a 21st Century Military, 1996. [Online]. Available:
Outstanding Performance Award. He is currently pursuing http://oai.dtic.mil/
a Master of Science degree in Aeronautics and Astronautics [9] E. Bonabeau, “Agent-Based Modeling : Methods
Engineering at Purdue University. His primary research area and Techniques for Simulating Human Systems Eric
is in Aerospace Systems; specifically, appraising the impact Bonabeau Proceedings of the National Academy of
of airline operations and competition on future environmental Sciences of the United States of America , Vol . 99 , No
emissions. . 10 , Supplement 3 : Arthur M . Sackler Colloquium
of the National Ac,” in National Academy of Sciences of
the United States of America, vol. 99, no. 10, 2002, pp.
7280–7287.
[10] R. Axtell, “Why agents?: on the varied motivations
Parth Shah is currently pursuing a for agent computing in the social sciences,” Center
Masters of Science at Purdue Univer- on Social and Economics Dynamics - The Brookings
sity under Professor Daniel DeLauren- Institution, no. 17, pp. 1–23, 2000. [Online].
tis in system-of-systems. His research Available: http://www.brookings.edu/{∼}/media/
interests are architecture asset selection research/files/reports/2000/11/technologyaxtell/agents.
using portfolio optimization and regres- pdf$\delimiter”026E30F$nhttp://www.brook.edu/{∼}
sion techniques, agent-based modeling, /media/Files/rc/reports/2000/11technology{ }axtell/
and system-of-systems applications in agents.pdf
the defense domain. He graduated from [11] U. S. DoD, “DoD Architecture Framework Volume 1:
Purdue University in May 2014 with a Definitions and Guidelines,” Departmnet of Defense,
Bachelors of Science in Aeronautical/Astronautical Engi- Tech. Rep., 2007.
neering. He has held several internships in industry with GE [12] S. F. Railsback, S. L. Lytinen, and S. K. Jackson,
Aviation, The Boeing Company, and NASA. “Agent-based Simulation Platforms: Review and Devel-

8
opment Recommendations,” Simulation, vol. 82, no. 9,
pp. 609–623, 2006.
[13] D. A. DeLaurentis, “System-of-Systems Laboratory.”
[Online]. Available: https://engineering.purdue.edu/
people/daniel.a.delaurentis.1/html/SoSL.html
[14] D. N. Fry, R. Campbell, and D. A. DeLaurentis,
“Modeling Systems-of-Systems from Multiple Design
Perspectives: Agents, Interfaces, and Architectures,”
AIAA Modeling and Simulation Technologies Confer-
ence, vol. 1, pp. 1–15, 2015.
[15] “Sizing up the beast: The MH-53E Sea
Dragon in action,” jan 2014. [Online]. Available:
http://ic.galegroup.com/ic/ovic/NewsDetailsPage/
NewsDetailsWindow?failOverType={&}query=
{&}prodId=OVIC{&}windowstate=normal{&}
contentModules={&}display-query={&}mode=
view{&}displayGroupName=News{&}limiter=
{&}currPage={&}disableHighlighting=false{&}
displayGroups={&}sortBy={&}search
[16] S. B. T. D. E. Galatowitsch, “Undersea
mines grow smarter and deadlier,” vol. 23,
no. 3, pp. 57+, jan 1991. [Online]. Available:
http://go.galegroup.com.ezproxy.lib.purdue.edu/ps/
i.do?id=GALE{%}7CA10506741{&}v=2.1{&}u=
purdue{ }main{&}it=r{&}p=PPMI{&}sw=w{&}
asid=de7b1148d143a99cf2a0eeedebb9367f
[17] UK MoD, “Falklands 25: Background Briefing,”
1997. [Online]. Available: http://webarchive.
nationalarchives.gov.uk/20100503141944/http:
//www.mod.uk/DefenceInternet/FactSheets/
Falklands25BackgroundBriefing.htm

You might also like