The current national security environment and fiscal tightening make
it necessary for the Department of Defense to transition away from a
threat based acquisition mindset towards a capability based approach
to acquire portfolios of systems. This requires that groups of
interdependent systems must regularly interact and work together as
systems of systems to deliver desired capabilities. Technological
advances, especially in the areas of electronics, computing, and
communications also means that these systems of systems are tightly
integrated and more complex to acquire, operate, and manage. In
response to this, the Department of Defense has turned to system
architecting principles along with capability based analysis. However,
because of the diversity of the systems, technologies, and
organizations involved in creating a system of systems, the design
space of architecture alternatives is discrete and highly
non-linear. The design space is also very large due to the hundreds of
systems that can be used, the numerous variations in the way systems
can be employed and operated, and also the thousands of tasks that are
often required to fulfill a capability. This makes it very difficult
to fully explore the design space. As a result, capability based
analysis of system of systems architectures often only considers a
small number of alternatives. This places a severe limitation on the
development of capabilities that are necessary to address the needs of
the war fighter.
The research objective for this manuscript is to develop a Rapid
Architecture Alternative Modeling (RAAM) methodology to enable
traceable Pre-Milestone A decision making during the conceptual phase
of design of a system of systems. Rather than following current trends
that place an emphasis on adding more analysis which tends to increase
the complexity of the decision making problem, RAAM improves on
current methods by reducing both runtime and model creation
complexity. RAAM draws upon principles from computer science, system
architecting, and domain specific languages to enable the automatic
generation and evaluation of architecture alternatives. For example,
both mission dependent and mission independent metrics are
considered. Mission dependent metrics are determined by the
performance of systems accomplishing a task, such as Probability of
Success. In contrast, mission independent metrics, such as
acquisition cost, are solely determined and influenced by the other
systems in the portfolio. RAAM also leverages advances in parallel
computing to significantly reduce runtime by defining executable
models that are readily amendable to parallelization. This allows the
use of cloud computing infrastructures such as Amazon's Elastic
Compute Cloud and the PASTEC cluster operated by the Georgia Institute
of Technology Research Institute (GTRI). Also, the amount of data that
can be generated when fully exploring the design space can quickly
exceed the typical capacity of computational resources at the
analyst's disposal. To counter this, specific algorithms and
techniques are employed. Streaming algorithms and recursive
architecture alternative evaluation algorithms are used that reduce
computer memory requirements. Lastly, a domain specific language is
created to provide a reduction in the computational time of executing
the system of systems models. A domain specific language is a small,
usually declarative language that offers expressive power focused on a
particular problem domain by establishing an effective means to
communicate the semantics from the RAAM framework. These techniques
make it possible to include diverse multi-metric models within the
RAAM framework in addition to system and operational level trades.
A canonical example was used to explore the uses of the
methodology. The canonical example contains all of the features of a
full system of systems architecture analysis study but uses fewer
tasks and systems. Using RAAM with the canonical example it was
possible to consider both system and operational level trades in the
same analysis. Once the methodology had been tested with the canonical
example, a Suppression of Enemy Air Defenses (SEAD) capability model
was developed. Due to the sensitive nature of analyses on that
subject, notional data was developed. The notional data has similar
trends and properties to realistic Suppression of Enemy Air Defenses
data. RAAM was shown to be traceable and provided a mechanism for a
unified treatment of a variety of metrics. The SEAD capability model
demonstrated lower computer runtimes and reduced model creation
complexity as compared to methods currently in use. To determine the
usefulness of the implementation of the methodology on current
computing hardware, RAAM was tested with system of system architecture
studies of different sizes. This was necessary since system of systems
may be called upon to accomplish thousands of tasks. It has been
clearly demonstrated that RAAM is able to enumerate and evaluate the
types of large, complex design spaces usually encountered in
capability based design, oftentimes providing the ability to
efficiently search the entire decision space. The core algorithms for
generation and evaluation of alternatives scale linearly with expected
problem sizes. The SEAD capability model outputs prompted the
discovery a new issue, the data storage and manipulation requirements
for an analysis. Two strategies were developed to counter large data
sizes, the use of portfolio views and top `n' analysis. This proved
the usefulness of the RAAM framework and methodology during
Pre-Milestone A capability based analysis.
| Identifer | oai:union.ndltd.org:GATECH/oai:smartech.gatech.edu:1853/43697 |
| Date | 04 April 2012 |
| Creators | Iacobucci, Joseph Vincent |
| Publisher | Georgia Institute of Technology |
| Source Sets | Georgia Tech Electronic Thesis and Dissertation Archive |
| Detected Language | English |
| Type | Dissertation |
Page generated in 0.0027 seconds