Modeling & Simulation Trends: Identifying Promising Contenders
Invited Workshop
Sponsorship
Boeing Phantom Works
Place & Time
Memorial Student Union, Ventana
C
Organizers:
H.S. Sarjoughian, ACIMS,
CSE,
ASU,
D.C.
Gross, Boeing Phantom Works,
W.V.
Tucker, Boeing Phantom Works,
Foundations
`04: A Workshop for VV&A in the 21st Century
https://www.dmso.mil/public/transition/vva/
http://www.scs.org/confernc/foundations/foundations04.htm
Purpose
The focus of the event is on existing and future trends in simulation
technology where researchers address issues related to simulation development
platforms and advancements in simulation approaches to enable rapid, credible
simulation developments and experiments.
Synopsis
Simulation has become a
cornerstone for a wide range of activities including engineering of systems.
While there is a widespread agreement on the role of modeling and simulation
concepts and methodologies as a basis for developing simulation technologies,
there exists a growing concern about their completeness and applicability in
new areas of enterprise-level systems on the horizon. Specifically, system of
systems with high degree of complexity and scale pose new challenges to
researchers and practitioners in the forthcoming decade.
The state-of-the-art
validation, verification, and accreditation concepts, theories, and practices
are believed to be only partially useful toward development of complex systems
and those that have intricate interactions with humans and natural systems. In
particular, present simulation technologies cannot support the study of
partially defined systems or well-defined systems that may be placed in
unanticipated situations in a systematic fashion. Furthermore as
simulation-based engineering enterprises continue to grow, it is imperative to
develop new concepts and techniques within and across traditional and
contemporary disciplines such as software and systems engineering, grid
computing, bio-informatics, and context-based simulation.
This event, therefore, is
aimed at investigating contemporary and extending traditional simulation
technology from a number of promising research directions including model
composability, cognitive-oriented simulation, incremental model specifications
and consistency checking, and simulation architectures quality attributes such
as scalability, usability, reconfigureability, and reliability. The panel will
discuss promising and state-of-the-art simulation technology while critiquing
their potential pitfalls and/or major gaps or shortcomings of today's outlook
for 2015.
Panelists will present 10 minute presentations. The remainder of the session will be devoted to discussing some key existing trends and hypotheses about those that can bring about or influence paradigm shifts in M&S for 2015. The invited session is comprised of 5 panelists and 10 participants. Participants are from academia and organizations with emphasis on shaping the future of M&S.
This event is sponsored by the Boeing Company,
The objective is to review and discuss the existing and future trends which can lead to new paradigms and/or revolutionizing M&S as it is known today. This finding of this investigation aided with research and the panel discussion will be made available to different parts of the M&S community as this document and the Journal of Defense Modeling & Simulation.
The following issues focus on some of the traditional and contemporary challenges facing modeling and simulation
o Conceptual and theoretical model composability
o Context-based customization of simulation models
o Technological innovations for delivering massive, high-fidelity simulation exercises
o Taming structural and behavioral complexity with multi-dimensional visualization
The above issues drive autocatalytic evolution and growth of modeling and simulation in the short, medium, and long-term.
o Process-based, automatic mapping of domain knowledge to simulation models (users, customers, and developers of simulations).
o Rise of complexity and scale: engineering, physical and engineering domains vs. M&S technology (new ways of modeling emerging from contemporary domains such as biological sciences and bio-informatics).
o Rise or fall of M&S as a discipline (workforce, M&S as science and technology vs. a tool).
o Sociological and economical impact (simulation studies as benchmark for making strategic decisions in an incremental fashion – creating a chain of high-order strategic decisions derived from low-order tactical decisions).
Gordon, Steve, Georgia Tech Research Institute,
Ghosh, Sumit*, Stevens Institute of Technology,
Gross, David, Boeing
Phantom Works,
Hild, Daryl, MITRE,
Mosterman, Pieter*,
MathWorks,
Paredis, Christiaan*,
Georgia Tech,
Sarjoughian,
Hessam**, Arizona State University,
Stevenson, Steve, Clemson University,
Tucker, William, Boeing
Phantom Works,
Vangheluwe, Hans*, McGill University,
Wilsey,
Phil*, University of Cincinnati,
* Panelist
** Moderator
1. ACIMS, (2003), “Second Workshop on Ultra Large Networks: New Research Directions in Modeling and Simulation-based Security,” http://www.acims.arizona.edu/EVENTS/ ULN03/ULN03MainPage.htm.
2. Banks, J. (2001), “Future of simulation,” Panel, WSC, Pages: 1453–1460.
3. Jain, S. (1999), “Simulation in the next millennium,” WSC, p. 1478–1484.
4. Kuljis J. and R. Paul, (2000), “A review of web based simulation: whither we wander?”, WSC, p. 1872-1881.
5.
Lunceford, W.H. and E.H. Page, (2002), “Grand
Challenges for Modeling and Simulation,” Editors, Western Multiconference,
6.
7. McQuay, W.K. (1997), “Air Force Modeling & Simulation Trends,” Program Manager, Sept/Oct, p. 128-132.
8. Nicol, D., et al., (1999), “Strategic directions in simulation research,” panel, WSC, p. 1509-1520
9.
Tucker, W.V. and D.C. Gross (1997), “Simulation
Trends and Implications,“ Simulation Multiconference,
April,
Position Statement (Sumit Ghosh): This panelist agrees, emphatically and respectfully, with the workshop synopsis in that a system of systems with high degrees of complexity and scale pose new challenges to researchers and practitioners in the forthcoming decades. Based on current experience, this panelist further proposes that modeling and simulation will, out of sheer necessity, become integrated into the disciplines of law, business, biology, medicine, transportation, and space. Just as today's computers can no longer be designed by a single designer or even the combined manual efforts of a team of designers, many medicinal surgical procedures in the future are likely to require sophisticated robotics and secure networks working in unison or even hundreds of micro-miniature robotics executing concurrently to repair the brain. Before such procedures may be realized, the only mechanism through which they may be studied in scientific detail is modeling and simulation. This panelist submits that it is critical to pursue behavior modeling and asynchronous distributed simulation in both research and education. The growing number of users and computationally intelligent nodes, vastly increasing geographical distance of coverage, and fast scale of interactions imply that asynchronous hardware, asynchronous software, and asynchronous distributed algorithms are inevitable and the most natural answer to the challenges. A key proposal is to develop a new language, networked computational systems design language and execution environment (NCSDL) which will provide a framework not only to describe and simulate any complex system but also reason about their reliability, vulnerability, performance analysis, operational testing, and evolution. The idea of NCSDL parallels the effort of high-level hardware design languages in the 1980s, notably VHDL, whose goal had been to enable the systematic design of synchronous computer systems.
Bio-Sketch: Sumit
Ghosh is the Thomas E. Hattrick '42 Endowed Chair
Professor of Information Systems Engineering in the Department of Electrical
and Computer Engineering at Stevens Institute of Technology in
Position Statement (Pieter J. Mosterman) In the realm of embedded control system design,
there is a wide spectrum of models that are being used. On the one end, very
detailed models of reality are used. These models typically capture the
behavior of the implementation (be it in hardware or software). On the other
end of the spectrum are very abstract models that closely relate to the
conceptual functioning of the system. These models are abstract enough to apply
analysis and synthesis methods.
In the design of an embedded controller, two
parallel trajectories in opposite directions are traversed. The plant that is
to be controlled is modeled based on first principles and system identification
methods. The derived model is then simplified by abstracting implementation
detail, piecewise linearizing, and removing
superfluous detail. This ultimately leads to a model that is sufficiently
abstract to apply controller synthesis methods. The synthesis results in a
model of the controller and the next task is to augment this controller model
with increasing detail to arrive at a specification that can be mapped onto an
implementation.
Looking towards the next decade to inventarize the simulation challenges it beholds for the
embedded control system industry, we find different needs at either end of this
spectrum. On the one hand, there is a need
to handle execution complexity that is much beyond what simulation engines can
presently handle. On the other hand, there is a need to provide intuitive,
domain-specific, modeling formalisms, possibly generated dynamically.
The execution complexity derives from plant
models that include a high-level of realism and that are increasingly composed
from model fragments. This compositional paradigm is expected to become a prime
methodology in industry where models are going to be used as a type of electronic
data sheet. As such, it constitutes the basis for supporting data based
acquisition. Suppliers then provide the original equipment manufacturer with a
model of their products and this model can be immediately inserted into an
envisioned design to verify whether it satisfies the requirements. The
challenge here comes from the stupendous execution complexity that comes about
when simply composing models with arbitrary detail together. In particular,
there will be widely varying time scales of the behavior and many interacting
discontinuities that cannot be handled efficiently.
The challenge at the other end of the
spectrum is of a more conceptual nature. Here the goal is to facilitate a
communication between human and machine to explain the envisioned behavior,
based on which the machine derives a computational recipe for a desired
platform to generate this behavior. To make giving this explanation convenient,
a 'white-board' type approach becomes necessary, in which the human can
intuitively draw annotated schematics such as those used in inter-human
explanation, that are then interpreted by the machine.
These different challenges are related by
their model-based nature. They can be addressed by methods studied in the
emerging field of Computer Automated Multi-Paradigm Modeling (CAMPaM). CAMPaM
aims to provide a framework that supports the handling of multiple formalisms
and multiple levels of abstraction by exploiting meta-modeling and model
transformations. In CAMPaM, simulation is regarded a model transformation that
allows the explicit modeling of the simulation algorithms, which puts
simulation on an equal footing with model transformation. For example,
continuous-time simulation may become the transformation of a model given in a
differential equation formalism into a time/value(s) tuple
formalism. Thinking of simulation as a model transformation, and model
transformations in a general sense, they can be exploited to reduce the
simulation complexity of models. This may allow reducing the complexity that
arises from simply composing model fragments so that a variable-step solver can
efficiently simulate. Moreover, the reduction may be necessary in case
real-time simulation is requested. In this case, accuracy can be automatically
sacrificed for simulation speed.
CAMPaM also supports the 'on-the-fly' design
of modeling formalisms, such as often is the case in 'white-board' explanation.
When elaborating on the behavior of an annotated schematic, scenarios are
typically presented that could be emulated by a machine and modeled as
transformation, for example in terms of graph rewriting rules. These rewriting
rules may just change the state of a model, and not change the formalism, but
could even transform a model into one in a different formalism. For example, in
modeling a physical system, often an ideal picture diagram is drawn, from which
then the underlying physical phenomena such as dissipation and inertia are
derived.
As the linchpin in
addressing these challenges, CAMPaM takes an important place and its methods
should be made available through technologies that are sufficiently robust to
set out and meet the challenges sketched in this text.
Bio-Sketch:
Pieter Mosterman is a senior research scientist at The MathWorks, Inc. in
Dr. Mosterman is the invited session chair for the 2006 IEEE International Symposium on CACSD, co-chaired the 14th International Workshop on Principles of Diagnosis (2003), he has served on the program committee of several conferences and workshops, and he has organized special sessions at the 2004 IEEE International Symposium on CACSD the 2001 IEEE Conference on Control Applications, the 2000 IEEE International Symposium on CACSD, and Eurosim ’98. He is currently a member of the IFAC Technical Committee on CACSD, chair of the IEEE CSS Action Group on Hybrid Dynamic Systems for CACSD, editor of SIMULATION: Transactions of The Society for Modeling and Simulation International for the area of Mechatronics, and associate editor of IEEE Transactions on Control System Technology and the International Journal of Applied Intelligence. He was also guest editor of special issues of ACM Transactions on Modeling and Computer Simulation and IEEE Transactions on Control System Technology on the topic of CAMPaM.
Position Statement (Chris Paredis) In
this position statement, I consider modeling and simulation in the context of
engineering design. Continuous advances
in computing and networking capabilities are fundamentally changing the
discipline of engineering design. There
is abundant capacity to capture and store huge volumes of data about engineered
systems and processes; there is processing power to quickly perform complex
analyses and optimizations; and there is networking bandwidth to share large
datasets in real-time among distant collaborators. The challenge is to use all these
capabilities such that they improve the designer’s ability to make rational
decisions.
To support decisions, one needs to provide the appropriate supporting information quickly, accurately and economically. This leads to the following question: How should one discover, formalize, catalogue, retrieve and apply knowledge in an efficient fashion, resulting in accurate information in support of product lifecycle decisions? It is my perspective that information can be generated efficiently through the use of modular, composable, and reusable knowledge representations, and that this information can be expressed accurately through the use of novel, formal, more expressive representations and methods for uncertain knowledge and information.
Composable Models: The exact information needed to support a decision is rarely available directly without any further processing. Instead, one usually uses models (knowledge) to extract or compute the appropriate information. This is a two-step process: 1) Formulate the model, and 2) Solve the model numerically. With the continuous increases in computing power and improvements in simulation technology, the bottleneck in this two-step process is quickly shifting towards problem formulation rather than problem solution. To reduce the time needed for model formulation, the most promising approach seems to be to formalize simulation models in a modular, composable and reusable fashion. Model Composition is based on the characteristic that there exists a strong parallel between the composition of physical components into a system and the composition of the corresponding simulation models into a system-level simulation model. For this approach to work, composable model chunks need to have a well-defined interface through which they interact with each other. Much of the research in composable simulations has focused on low-level run-time interoperability. I would like to encourage the community to focus on defining model interfaces at a much higher level: What are the semantics of the interface? How can one characterize the applicability and validity of composable model chunks and of the resulting compositions? How can models be expressed in a declarative fashion such that composition can occur independently of procedural, run-time considerations?
Accurate representations for uncertain information and knowledge – In making design decisions, one has to recognize the fact that much of the information and knowledge on which the decisions are based is uncertain. If one does not take this uncertainty into account, one may arrive at flawed design decisions leading to defective products or costly redesign. Yet, in current design practice, many of the simulations and analyses are still deterministic, and uncertainty is handled only implicitly through safety factors. The question is: How can one explicitly account for uncertainty in design such that the representation is conservative (relies only on objective information) while still allowing one to process it efficiently? Certain types of uncertainty, such as manufacturing variability or weather conditions, are inherently random (aleatory uncertainty) and can be accurately represented using probability distributions. However, there is also uncertainty due to a lack of knowledge (epistemic uncertainty) that is very common in design—for instance, uncertainty due to design details that still remain to be decided. Representing such uncertainty using probability distributions would imply more knowledge than is objectively available—the representation would no longer be conservative. Instead, set-based intervals are best suited for epistemic uncertainty. Since, in design, both types of uncertainty co-exist, there is a need for novel mathematical formalisms that allow one to consider both types of uncertainty simultaneously. An example of such a formalism is Probability Bounds Analysis. I would like to encourage the community to investigate formalisms that allow for a more expressive representation of uncertainty, while still maintaining the ability to compute (simulate) efficiently with the uncertain information.
Bio-Sketch: Dr. Christiaan
J. J. Paredis is currently an Assistant Professor in the
Position Statement (Hans Vangheluwe) The complexity of systems we analyze and design, in particular those of an inter-disciplinary nature, has increased dramatically over the last few decades. This complexity is due to a number of factors such as:
§
the
number of components of the system;
Modeling and Simulation research in different application domains, on different formalisms, as well as on different techniques and tools has often been isolated. As a consequence, possibilities for cross-fertilization are lost and all too often sub-optimal solutions are produced due to “reinventing the wheel”.
In some domains such as mechatronics, attempts are made to unify and integrate the above to ease the analysis and design of truly multi-disciplinary, complex systems. The modeling and simulation viewpoint provides a general and intuitively appealing framework for unification.
In the following attempt to identify trends in Computer-based Modeling and Simulation Technology we follow the definition given by Tucker and Gross: a trend is an observable factor affecting business and/or technical decisions.
When looking at the standard platforms for communicating advances in Modeling and Simulation such as the Winter Simulation Conference, ACM's TOMACS and SCS's Simulation/Transactions, we observe a multitude of applications of modeling and simulation to problems of ever increasing complexity. However, the Grand Challenges of the past which led to the standard modeling and simulation approaches and more recently to standardizing simulator inter-operability, seem to all have reduced in both focus and intensity. Note that new research trends typically show up as a sigmoid function of publications versus time.
Actually, looking more closely at Modeling and Simulation publications as well as tools, the distinction between Modeling and Simulation on the one hand and applications and enabling (software and hardware) technology on the other hand seems to become less and less pronounced. This blurring of boundaries is in our opinion indicative of the widespread acceptance and adoption of Modeling and Simulation methods, techniques and tools. Modeling and Simulation have become main-stream. To support this claim, it suffices to look at the Software Engineering community (for example, at recent OOPSLA and ICSE conferences and in new journals such as Software and Systems Modeling (SoSyM) where Model Based – albeit not Simulation Based-- software development is rapidly becoming accepted as the new trend beyond object-oriented programming and design. Conversely, thanks to the increasing performance of off-the-shelf hardware and the standardization of software components, many modeling and simulation (technology) problems have now become problems which are addressed in the larger hardware and software design (and co-design) communities.
The above observations lead to the insight that what sets Modeling and Simulation apart is:
§ the aim to analyze and build truly complex systems (in every sense mentioned previously)
§ a particular, model-centered world view. It is noted that due to the recent blurring of the distinction between Modeling and Simulation on the one hand and generic Software Engineering on the other hand, this distinct model-centric view (possibly induced by the nature of typical applications such as hardware-in-the-loop simulation) has been somewhat diluted. As an example of the weakening of this model-centered world view, connecting federates in the HLA tends to neglect whether the models being simulated can be meaningfully combined.
Despite the convergence between Modeling and Simulation on the one hand and diverse research areas on the other hand, we are still hitting the complexity barrier in analysis and design of complex systems. One could argue that now is the time for a paradigm shift (in the sense of Kuhn) in the way we deal with complex systems.
The emerging trend, fuelled by both a pressing need and by the growing availability of appropriate theory and (software) technology is to ``model everything''. This means that no part of the structure nor behavior of the system under study (or to be built) must be left un-specified or implicit. This includes the development, deployment and maintenance process at all levels: workflow, scheduling, etc. Above all, this includes the explicit modeling of transformations between various modeled artifacts (as that is what the process achieves) ! A particular instance of this is the modeling, not only of the system-under-study, but also of modeling and simulation experiment including aspects such as model initialization, possibly recursively (simulating modeling and simulation experiments whose decision variables are in turn influenced by the results of modeling and simulation experiments). Note how a concrete instance of the “model everything” philosophy is found in Modeling and Simulation Based Acquisition. We believe that a visible effect of the “model everything” trend is the emerging field of Computer-Automated Multi-Paradigm Modeling (CAMPaM) which looks at past and future problem-solving (i) at different levels of abstraction, (ii) using multiple formalisms, by means of (iii) meta-modeling. Spanning these three orthogonal directions are model transformations. Current state-of-the-art model engineering uses meta-modeling and graph rewriting as concrete enabling technologies(as implemented in the author's AToM3, A Tool for Multi-formalism and meta-modeling and Vanderbilt's GME, the Graph Modeling Environment).
Meta-modeling is a crucial component of domain-specific modeling. The domain-specific modeling trend explicitly supports modeling and simulation (and animation, etc.) in the application domain, making underlying technology completely transparent. It is noted that, though powerful, modeling in the Unified Modeling Language (UML) is still in the software (technology) realm, not in the domain of interest. Domain-specific modeling can obviously only work if the transformation of domain-specific models to the solution/implementation domain can be automated. The first step in automation is to model the transformation, so again transformation takes a central place. Note how ideally, every single part of a system will be synthesized from models (about which we can reason, which can be re-used, ...).
Of course, for a “model everything” approach to work, we must navigate correctly between different levels of abstraction. In an engineering context, we may enforce moving between levels of abstraction “by construction” (using domain-specific formalisms). In modeling of real-world existing systems such as biological systems, the problem of moving between levels of abstraction is much more of a challenge -- it's science, not engineering-- (though even here, we may enforce some amount of modularity when engineering systems such as bio-activated Waste Water Treatment Plants).
This brings us to a crucial point: the need for modular composition (and re-use) of models. A distinction needs to be made between simulator inter-operability on the one hand and meaningful coupling of models (brought to life by simulators) on the other hand. By far not enough attention is paid to the ``meaning'' of coupling models. Notions of semantic expressiveness such as compositionality, bi-similarity, and full abstraction all tell us something about how well we can ``plug'' models into different contexts and re-use them without changing their meaning. New formalisms are being designed which exhibit these desirable properties which should not be taken for granted. For example, block diagrams (as used in The Mathworks' Simulink) are often assumed to be modular. One needs to look at behavior (and not just syntax) however. There is a lot of work to be done on designing formalisms (and subsequently supporting modeling and simulation tools) with appropriate modularity. Note that under certain circumstances, we do not want modularity, because it's not needed, of because of performance constraints (at least at the simulation level - as opposed to at the modeling level). We postulate that an appropriate process, building on theory and using appropriate tools, can enforce “by construction” the production of systems of higher quality, thereby making verification and validation easier. Note that a correct transformer for example will greatly reduce the verification effort (for example, by automatically generating test vectors).
A more long-term trend in our opinion is the use of sensor networks and more generally of ubiquitous computing (as recently introduced at MIT). Here, collections of interacting agents with relatively simple individual behaviors will lead, at a macroscopic level to emergent behavior. Such agent networks may either be simulated or actually deployed in hardware. They may be used for the modeling and simulation based analysis of systems about which typically not much is known: economic systems, group dynamics, biological systems, etc. It is highly desirable to not only to analyze, but also to design/engineer agent networks to implement, at the emergent behavior level, system requirements. Future research will show whether this is realistic.
In conclusion, we believe that the new trends in Modeling and Simulation and the underlying paradigm shift towards thinking of “everything as a mode” (with the accompanying shift from simulation-focus to modeling-focus) will put Modeling and Simulation in a central place in the effort to push the boundaries of complexity.
Bio-Sketch: Hans
Vangheluwe holds degrees in Theoretical Physics, Education, Computer Science
and Science (D.Sc.) from
He has been a Research Fellow at the Centre de Recherche Informatique de Montreal (CRIM), Canada; the Concurrent Engineering Research Center (CERC) at WVU, Morgantown, WV, USA; at the Delft University of Technology, The Netherlands; and at the Supercomputing and Education Research Center (SERC) of the Indian Institute of Science (IISc) in Bangalore, India.
He was the co-founder and coordinator of the European
Union's ESPRIT Basic Research Working Group 8467 “Simulation in
At
More recently, he has developed (with Prof. Juan de Lara), AToM3, A Tool for Multi-formalism and Meta-Modeling. AToM3 combines meta-modeling and graph rewriting to allow for rapid development of domain-specific modeling, simulation and synthesis environments. AToM3 is a vehicle for research into Computer Automated Multi-Paradigm Modeling (CAMPaM), a concept combining multi-formalism, multi-abstraction and meta-modeling introduced by Prof. Vangheluwe and Dr. Mosterman to tackle problems of ever increasing complexity.
Position Statement (Philip A. Wilsey) Simulation will continue to play an increasingly
important role in the design and analysis of complex systems. More importantly, we will see an increased
reliance on simulation results to guide major choices involving huge economic
commitments by those making the choice.
Key areas for simulation such as performance, interoperability,
verification, and validation are well-known and ongoing challenges that must be
aggressively studied and advanced. In
addition to these problems, we must also focus on model construction, support
for analyzing simulation results, and management of complex software
environments. Well-known software
development techniques such as interface assertions, test-first design, and
mock objects need to become part of the modelers mind-set. Tools and techniques for improved output
analysis should also be pursued. Ideally
formal methods would be incorporated (although their adoption by the software
engineering community remains unrealized).
Tool interoperability will also become more important. Finally, as infrastructure support for model
construction expands, we must learn how to manage tool complexity so that they
are useful to the average modeler. This
is already a problem in many tools; without previous experience or classroom
training, tools like gimp, quanta, eclipse, rational rose, and so on can be
nearly impossible to use.
Bio-Sketch:
Philip A. Wilsey is a Professor of Computer Engineering at the