1.1 ADS Architecture Design and Evaluation (ADS-ADE) Task 2 - 18

Coordinator: Ratan K. Guha

1.2 ADS Visualization and Synthetic Environment (ADS-VSE) Task 19 - 32

Coordinator: Patrick Bobbie

1.3 ADS Knowledge Base Systems (ADS-KBS) Task 33 - 48

Coordinator: Richard Aló

1.4 Student Outreach and Training Program Task 49 - 53

Coordinator: Muddapu Balaram

2.1 ADS Architecture Design and Evaluation (ADS-ADE) Task 55 - 57

Coordinator: Ratan K. Guha

2.2 ADS Visualization and Synthetic Environment(ADS-VSE) Task 58 - 59

Coordinator: Patrick Bobbie

2.3 ADS Knowledge Base Systems (ADS-KBS) Task 60 - 62

Coordinator: Richard Aló

2.4 Student Outreach and Training Program (ADS-KBS) Task 63 - 68

Coordinator: Muddapu Balaram

3.1 Visits by ADSRC Researchers and Students 70

Appendix: Publication Abstracts 71 - 89

The main objectives of the research performed by the ADS Architecture Design and Evaluation team are: (1) to develop techniques to model/evaluate the performance of large scale ADS systems, (2) to develop algorithms, based on DoD High Level Architecture (HLA), for improving the scalability and reducing the bandwidth requirement of distributed simulation systems, and (3) to improve solution techniques used by tools for the modeling and performance evaluation of DIS systems.

Specifically, our objectives in the general area of research for HLA are to (1) examine various approaches to relevance filtering in the context of the HLA and devise alternative algorithms that can improve their execution efficiency and filtering efficiency, (2) develop better performance analysis methodologies to fully determine the performance characteristics of the various data distribution approaches, (3) propose changes to current HLA specifications to incorporate objects fidelity requirements in its data distribution management services, and (4) investigate methods for efficiently implementing the new services.

Regarding implementation issues for large distributed systems, our objectives of research are to (1) develop fundamental concepts in distributed mutual exclusion for efficiency and fault tolerance capabilities of the system, and (2) design schemes for channel assignments and handoff handling in mobile cellular networks.

For general purpose tools, our objective of research is to develop easily accessible interval computation software which can reliably perform fundamental interval arithmetic, set operation and bound elementary interval functions.

The research on filtering and routing spaces is crucial to the scalability of HLA currently being developed under the auspices of the Defense Modeling and Simulation Office as the DoD-wide standard for simulation. The results of this research have provided good insight into the behavior of relevance filtering schemes, their real-time performance and their reliability. With this insight, the development of reliable and efficient filtering methods is greatly enhanced.

Wireless and mobile communication technologies will continue to play increasingly important role in the various functions of armed forces. The development of efficient channel assignment and handoff protocols is of one of the critical problems facing the designers of third generation wireless systems. Our research addresses the real-time performance issues of handoff protocols and seeks to develop efficient channel assignment and handoff schemes suitable for the delivery of UAV-collected imagery to/from mobile tactical vehicles and the delivery of location-based video/audio data to mobile users. Efficient handoff designs will also help the effort to incorporate live vehicles in military simulation training exercises.

Several teams have been formed to work on several sub-tasks. Areas of accomplishments include relevance filtering for large-scale simulation exercises, HLA data distribution management, distributed object standards, real-time protocols for channel assignments and handoff handling in mobile cellular environments, triangular level quorums for distributed mutual exclusion, an online interval calculator, and a parallel global optimization software tool.

The planned activities of each ADS-ADE sub-task are described in the individual reports given on the following pages.

The HLA, designated as the standard architecture for distributed simulation systems, has been supported by DMSO since its inception. Efficient and effective data distribution within this standard architecture is crucial to its scalability and future applicability. This project addresses issues that directly impacts the efficiency of data distribution in HLA.

The following tasks have been accomplished

1. Developed simulation programs in C/C++ to investigate various approaches to relevance filtering in distributed simulation systems.
2. Devised new approaches to that promise to improve the performance of data distribution management of High Level Architecture (HLA) distributed simulations. These include a hierarchical grid-based approach and a dynamic/variable resolution grid based approach.

    

3. Evaluated the performance of the new approaches under various simulation parameters and compared it to the traditional fixed-grid based approach that is employed in current implementations of the HLA.

    

We plan to expand our work to the following areas:

 

1. Investigate non-gridded approaches. In particular, we plan to investigate region clustering techniques. Region clustering is based on grouping objects based on the proximity of their interest regions in a routing space. Region clustering techniques have the potential of achieving near optimal filtering performance. However, their efficiency, in terms of overhead cost, and scalability remain questionable and need to be further investigated.

    

    

2. Extend the hierarchical approach to incorporate non-gridded algorithms. In particular, the interfaces between the different levels of the hierarchy need to be further developed to provide correct and efficient interfacing between the "sub-federates" of the hierarchy.

    

    

3. Develop better performance analysis methodologies to fully determine the performance characteristics of the various data distribution approaches. In our previous studies, synthetics simulation was used to evaluate different filtering algorithms. We plan to use trace-driven simulation based on traces generated from actual scenario simulations.

    

Wireless and mobile communication technologies will continue to play increasingly important role in the various functions of armed forces. The development of efficient channel assignment and handoff protocols is of one of the critical problems facing the designers of third generation wireless systems. Our research addresses the real-time performance issues of handoff protocols and seeks to develop efficient channel assignment and handoff schemes suitable for the delivery of UAV-collected imagery to/from mobile tactical vehicles and the delivery of location-based video/audio data to mobile users. Efficient handoff designs will also help the effort to incorporate live vehicles in military simulation training exercises.

M = M1 È M2 È M3

To simplify the discussion, we shall assume that the three sublists have equal size, i.e., |M|=3m for some integer m. Let 2 denote the unary operator that reverses the order of elements in an ordered list. Thus we have

M1 = {1, 2, …, m} 2 M1 = {m, m-1, …, 1}

M2 = {m+1,m+2, …, 2m} 2 M2 = {2m,2m-1, …, m+1}

M3 = {2m+1, 2m+2, …, 3m} 2 M3 = {3m, 3m-1, …, 2m+1}

The three sublists are assigned as the nominal sets of channels for successive cells in alternating fashion. Each sublist is toggled using the 2 operator after each assignment. For example cell K-3, cell K, and cell K+3 all have the same set of nominal channels assigned to them but the perceived order of these channels in the three cells are M1, 2 M1 and M1, respectively. Similarly, cell K-2 is assigned M2, cell K+1 is assigned 2 M2, and cell K+4 is assigned M2.

 

2. The scheme has a communication cost of 2 messages per channel request or channel release.

    

    

3. Assuming channels are encoded as integer indexes without gaps, the storage overhead of the scheme at each cell can be reduced to O(1) without affecting the time complexity.

    

We also plan to extend our research to the case of two-dimensional cellular networks. Our focus will be to develop predictive channel reservation for mobile cellular networks based on GPS measurements. Using GPS and dead-reckoning, each mobile can trace its path and report its movement to its current base station. This base station extrapolates the path of the mobile and initiates a channel reservation request to the neighboring cell that the mobile is heading to.

An HLA routing space is a multidimensional coordinate system by which federates can implement a filtering mechanism without requiring the RTI to have complex domain specific knowledge. The three components of a vector location are used as the three variables of a three-dimensional routing space. We have used a circular region of interest to determine the relevancy of transmitted messages and implemented filtering algorithms at the sending receiving gateways. Both analysis and simulation have been used to study the performance and reliability of relevance filtering algorithms.

Filtering at transmission and filtering at reception are two important requirements in all DIS exercises. Algorithms for distance based filtering and grid based filtering was developed and used in local and gateway dead-reckoning. There are other approaches to filter the irrelevant messages. One of the methods is calculating the fitness of the entity using genetic algorithm approach. The approach provides better results because relevance of the message normally depends upon type of entity, position of entity, direction it moves, and other characteristics. The total characteristics are coded into blocks of bits and test for fitness. The messages are delivered to the best-fit entities. The GA method is faster and more accurate. Some work has been done using GA approach. Other approaches in progress include:

 

• Mapping the entities located in the same terrain region to hosts located in the same LAN helps to localize the set of hosts that would need to receive state update messages from each entity.

   

   

• Grid based data distribution.

   

   

• Estimating the entity position using previous state and related information.

   

The group demonstrated progress and expected to accomplish in the following topics during the next two years.

 

• Two stage filtering scheme by adapting it to the routing space mechanism of HLA, investigating the impact of multicast algorithms on its performance and performing reliability analysis of filtering in HLA.

   

   

• Grid-based data distribution schemes in the HLA and comparison to other methods.

   

   

• Circular cell approach to calculate relevance of the entity in the gateway (federation).

   

   

• The GA methodology includes building blocks approach to find the fitness of each entity of gateways and filters appropriately. Genetic algorithm has built in parallel search property. The processing is done much faster and more accurately. Progress is now towards classifier systems to predict the best-fit entities at a given time. The algorithms will be extended to distribute the messages to selected federates. Since parallelism is built in property of GA and other approaches can be coded to GA parameters, the GA approach will be more popular for RF.

   

The HLA, designated as the standard architecture for distributed simulation systems, can only be implemented as a large distributed system. For a system like this, it is desirable to have the quorum size as small as possible to enter a critical section. On the other hand, to increase the fault tolerance capability, the coterie size should be as large as possible. This project addresses issues that influences the efficiency and fault tolerance capabilities in the implementation of HLA.

We have developed the concept of triangular level quorums as follows. Let U = {1, 2, …, N} be the nodes in a distributed system. Let the m sets S₁, S₂, …, S_m be subsets of U such that S_i Ç S_j = Æ for 1 £ i ¹ j £ m, | S₁ | = 1, | S_i | = | S_i-1 | + 1 for 2 £ i £ m, and U = S₁ È S₂ È … È S_m. The m-level structure S₁, S₂, …, S_m forms an equilateral triangular structure, and the level coterie constructed from the triangular structure is called an m-level triangular level coterie, denoted as D _m. Because |S₁| = 1, and | S_i | = | S_i-1 | + 1 for i = 2, 3, …, m, we have | S_i | = i for i = 1, 2, …, m.

Let U={1, 2, 3, …, N} be the N nodes in a distributed system. The N nodes are logically arranged into m levels S₁, S₂, …, S_m, where S_i are non-empty subsets of U. S_i Ç S_j = Æ , for all 1 £ i ¹ j £ m, and U = S_{1 È} S₂ È …È S_m. If | S_i | = i for 1 £ i £ m, then the ND level coterie D _m = Q₁ È Q₂ È … È Q_m is called an m-level triangular level coterie, q Î D _m is an m-level triangular level quorum, where Q_i, 1 £ i £ m, are defined as

m

Q₁ = {S₁ È ( È {X_j}) | X_j Î S_j, j = 2, 3, …, m}

j=2

m

Q₂ = {S₂ È ( È {X_j}) | X_j Î S_j, j = 3, 4, …, m}

j=3

…

m

Q_i = {S_i È ( È {X_j}) | X_j Î S_j, j = i+1, i+2, …, m}

j=i+1

…

Q_m = {S_m} 

An Example : let U = {a_ij | 1 £ i £ 5, 1 £ j £ i}, S_i = { a_ij | 1 £ j £ i}, 1 £ i £ 5. The nodes are logically arranged into a 5-level equilateral triangular structure as depicted in figure 2. By definition 3, the following are some instances of the 5-level triangular quorums constructed from S_i, i = 1, 2, ..., 5.

q₁ = {a₁₁, a₂₂, a₃₂, a₄₃, a₅₅}, q₂ = {a₂₁, a₂₂, a₃₁, a₄₃, a₅₄}, q₃ = {a₃₁, a₃₂, a₃₃, a₄₅, a₅₃},

q₄ = {a₄₁, a₄₂, a₄₃, a₄₄, a₅₂}, q₅ = {a₅₁, a₅₂, a₅₃, a₅₄, a₅₅}, q₆ = { a₁₁, a₂₁, a₃₁, a₄₁, a₅₁}.

It is clear that q_i Ç q_j ¹ Æ , and q_i Ë q_j for 1 £ i ¹ j £ 6.

Note that | q₁ | = | q₂ | = … = | q₆ | = 5.

Triangular Structure Level Size of Level

O 1 |S₁| = 1

O O 2 |S₂| = 2

O O O 3 |S₃| = 3

O O O O 4 |S₄| = 4

O O O O O 5 |S₅| = 5

Figure : A 5-level triangular structure to construct a 5-level triangular level coterie

From the m-level triangular structure and the definition of D _m, any q Î D _m, q Î Q_i, for some 1 £ i £ m, the size of q equals | S_i | + m - i = m. That is, any quorum q in an m-level triangular level coterie D _m has the same size | q | = m. The number of quorums in an m level triangular level coterie D _m is (m! + m!/2! + m!/3! + … + m!/m!), we have m! £ | D _m | < 2m!. Let the total number of nodes in an m level triangular structure be N, then N = m(m+1)/2. This implies that m = (2N - m)^1/2 = ½ [(8N+1)^1/2 –1] » (2N)^1/2.

The following tasks have been accomplished

 

1. Developed a new approach, called triangular level coterie, to generate relatively small and equal-sized quorums.

    

    

2. Proved that the generated coterie has many interesting properties, such as non-domination, equal-sized quorums, no corresponding vote assignment.

    

    

3. Proved that the quorum availability of triangular level coteries converges as the number of nodes increases and satisfies the complementary property.

    

We plan to expand our work to the following areas:

 

1. Investigate trapezoid level coteries. If the number of nodes is not m(m+1)/2 for some integer m, and equal-sized quorums are required, then we can form a trapezoid level coterie.

    

    

2. Investigate incomplete triangular level coteries. If the number of nodes is not m(m+1)/2 for some integer m, another approach is to add dummy nodes to form triangular level structure, and then to form triangular level quorums. The dummy nodes are then replaced by the real nodes.

    

M. Bassiouni, R. Guha, U. Vemulapati (UCF)

This project addresses the efficiency and effectiveness of data distribution in the HLA. Results of this project can improve the performance and scalability of large distributed simulations.

Research Objective and Significance:

The main objective of this project is to research efficient secure transmission methods. High volume of information is used to represent data for various computer transactions. Managing and/or securing these extremely large data sets can pose problems in storage space and processing time. Compression addresses the issues of manageability and encryption provides one of the most effective means of protection against common threats. The primary focus of this past year has been security.

Security is of paramount importance to the Army because of the sensitive nature of many documents. It is desirous to protect these documents from disclosure to unauthorized persons and modification by unauthorized persons. Compression is important because of the volume of documents. It is especially important to Distributive Interactive Simulation (DIS) because of the high volume of information (bits) in its various transactions. Many of the transactions, which involve images, can tolerate a fidelity loss without impeding their functionality. Thus, this leads to better real-time simulation with lessened storage space and processing time via compression.

Many encryption schemes were discovered through a literature search and more are being developed. The strength of any encryption scheme relies in its keys. The primary focus of this past year has been the McEliece public-key cryptosystem. This scheme is based on the theory of error correcting codes. In addition to its capability of offering secure transmission, it has been shown to possess the capability to correct communication channel errors. However, McEliece encryption scheme causes large data expansion.

Compression study was initiated as a possible method to counteract the data expansion due to McEliece encryption scheme. A technical literature search has indicated many research activities in such areas as data (text), image, and video. Currently, we have begun our investigation by reviewing digital signal processing principles, such as the FFT. We will also investigate other mathematical techniques such as wavelets.

Some "weak" keys have been identified which do not offer effective secure transmission. This has been incorporated into a key generation algorithm to safeguard against currently known "weak" keys. Also, a compression web-based tutorial was initiated to further enhance the knowledge of faculty, to develop a testbed of compression routines, and to ultimately introduce concepts to students.

The objective of this project is to develop an internet accessible interval computation tool which can reliably perform fundamental interval arithmetic, set operation and bound elementary interval functions with an interactive graphical user interface. Interval computation is a relatively new technology. It can better model the real world, and provide reliable solution bounds for application problems. Most people are still not familiar with interval computing. Interval calculator is not available to most users. This project provides an easy accessible fundamental computation tool for people who interested in interval computation.

Interval computing technology, as a basic computation method, can be applied to almost all computations required by Army applications.

To make the software package internet accessible and platform independent, we implemented it as a Java applet. The language we used is JDK 1.1.6. We also used Microsoft Visual J++ 1.1 to implement the graphical user interface.

The entire package contains the following classes:

 

1. Rmath Class: This class provides methods to perform directed rounding.

    

    

2. RealInterval Class: This class defines the interval type for real intervals.

    

    

3. IAMath Class: This class contains methods to perform basic interval arithmetic and set operations, and bound interval elementary functions.

    

    

4. IAException Class: This class handles error messages.

    

    

5. Cacpad and DisplayLayout classes: These two classes are for the interactive graphical user interface.

    

    

6. I_Calculator Class: This class is the main program which integrates all of the above to form a functional online interval calculator.

    

The initial implementation of the interval calculator is now available through the internet. To access it, users may visit the URL: http://gauss.dt.uh.edu/IntCalculator/I_C.html. A draft of the user guide can be found in a link in the above URL as well.

 

1. We will revise the implementations of some elementary functions by using true interval algorithm rather than using Java provided floating point methods.

    

    

2. We will add more features to the online calculator.

    

The objective of this project is to improve the performance of a generic non-linear global optimization interval software package named GlobSol through parallel/distributed computing. Non-linear global optimization is a fundamental research problem. With interval computing, GlobSol can reliably find global optimum in a given domain for many application problems. Our initial parallel implementation has significantly improved the performance of this package.

ParaGlobSol is a generic global optimization package. It can be used to solve optimization problems in the Army.

 

1. Our initial parallel implementations were done on networked Sun Ultra workstations.

    

    

2. We use MPI (the Message Passing Interface) to distribute the workload to multiprocessors.

    

    

3. By using the backward compatibility of Fortran 90, we have developed an MPI working environment that supports interval computations with Fortran 90. The original package GlobSol was written in Fortran 90. However, MPI does not have a Fortran 90 implementation.

    

    

4. Our implementation is coarse-grained SPMD (Single Program Multiple Data).

    

    

5. Both static and dynamic load-balancing techniques are applied to improve the overall efficiency.

    

    

6. Super-linear speedup was observed for some test problems.

    

 

1. Develop and Implement other parallel schemes

    

    

2. Improve overall efficiency and scalability of this parallel software package

    

    

3. Apply this package to solve some computation intensive Army related optimization problems.

    

The main objectives of the research performed by the ADS Visualization and Synthetic Environment team are: (a) To develop and evaluate algorithms for graphics and visualization; (b) To analyze the real-time performance of the algorithms; and (c) To tailor the algorithms to phenomenological (e.g., terrain, fire, smoke) datasets and distributed computations in synthetic environments using interactive visualization methods. Toward this end, the ADS-VSE team embarked on the following subtasks:

(1) A Computational Geometry Tutorial (CGT) system was developed to provide a distance learning capability for ADSRC-wide students and faculty in the study of basic concepts of computational geometry. The CGT system is continually enhanced to support curricular-oriented materials in computer graphics and visualization. The enhancement includes the introduction of Virtual Reality (VR) modeling component for animation, quality rendering, and visualization of graphic objects.

(2) A subproject focuses on the two-dimensional Ising model to represent and simulate the dynamics and behavior of fire and smoke in synthetic environments. Generally, we consider the elements of fire and smoke as a system of particles which undergo phase transitions, energy-level changes, and exhibit positional changes to accurately represent real-world phenomenon. However, real-time simulations of large systems with large two dimensional or three dimensional lattices require significant processing power. One thrust is to increase the performance of the simulation system by harnessing the overall computing power of the underlying distributed processors. The main goal is to improve the real-time performance of the Ising model using multiple processors. This subtask complements another Ising-related effort which focuses on the mathematical underpinnings and the physics of the model.

(3) In furtherance, a subproject is implemented for the design, development and implementation of efficient computer simulation of fire propagation, by extending the percolation model. The percolation approach has a well documented history in basic fire propagation behaviors through random media with the added effects of wind. An extension of this basic model to include topology, non-random media, moisture content and effects of both ground propagation and tree canopy propagation allows for realistic simulations of battlefield fire propagation.

(4) Another task focuses on providing a compendium of the field of fire and smoke modeling. The compendium also reports on the current physics-based modeling of flame propagation as well as flame spreading in fires, design and operation of gas-fired heat transfer equipment (combustors), and flame spreading over the surface of thermoplastic fuels during the ignition process in hybrid rocket motors.

(5) A subtask focuses on spoken language understanding for battlefield simulations. This work involves studying existing techniques for incorporating speech interactions in simulations. A spoken dialogue system allows a user to interact with a computer application using conversational speech. Web-based tutorial system for this speech-based subproject is being developed to demonstrate the applicability of spoken language-based operators into battlefield simulations and synthetic environments. The ModSAF and NPSNet environments are used to test these speech-based techniques.

(6) High volume of information is used to represent various computer transactions. Managing these extremely large data sets can pose problemsin storage space and processing time. The use of compression techniques to address these issues is investigated under another subproject.

(7) Another task focuses on the transport processes associated with phase change such as boiling (nucleate, film, and flow) which are the most efficient heat transfer modes that are characterized by small temperature differences and high heat fluxes. This effort dovetails with the Ising model for simulating phenomenology in virtual environments.

The CGT system is a useful resource for distance learning and offers a tutorial environment for learning computation geometry concepts outside of classroom setting. Its continual development to benefit all ADSRC participants is important, as it minimizes the learning curve and offers a quicker transition to the development and implementation of more advanced research topics in ADS.

Applying parallel and distributed computing methods to the Ising-based simulation allows the minimization of the overall simulation time. This step is important as it leads to near real-time visualization of geometric objects and entities in synthetic environemnts, e.g., virtual battlefield situations. The incorporation of an extended percolation model for fire simulation makes virtual environments, e.g., battlefield simulation, much more realistic. This improves the training of soldiers and facilitates their acquisition of military skills, therefore, enhancing their readiness for battle. Simulation models that represent battlefield situations also help to improve the understanding of phenomenological occurrences and their impact on the soldier.

Distributed simulation environments have played a significant role in an ever-increasing number of critical military and commercial applications. Applications ranging from fire safety training to virtual warfare benefit tremendously from this emerging technology. The virtual battlefield necessitates the inclusion of real-time obscurants such as fire and smoke into the computer simulation in complex ways. These models and simulations will be of wide relevance to government and industry organizations.

The incorporation of spoken language-based operatives in battlefield simulations provides a way to manage multiple participants, e.g., commanding officers or soldiers under training, who are interacting with a simulation using speech.

Simulation systems like DIS use a high volume of information to represent images in its various transactions. The transport of such data in DIS and virtual environments requires highly compressed data with little or no loss of information. The investigation in data compression and image understanding bears on the Army's continual need for maintaining high fidelity data and functionality.

The areas of accomplishments have been categorized into 'Research': related to impact on student and faculty research; 'Tools': related to software tools resulting from the ADS-VSE team; 'Student/Labs': related to impact on student research, training, and advancement to graduate studies; and 'Impact and Visibility': related to the overall impact of the ADSRC VSE effort.

Research: In addition to conference presentations, the ADSRC ADS-VSE team published several articles in both conference proceedings and journals. The listing of these results are summarized in the abstracts section (3), along with the faculty and students who participated in the development of these results in section (2) of this annual report. Further the summary pages in section (1) of this report include subsections on accomplishements which detail each task's results.

Tools: The CGT system, a graphics and visualization tutorial system, a spoken language-based dialogue system, an AI-based (fuzzy control) graphics system, simulated fire/smoke models for demonstrating the effect of fire propagation, and the MPI implementation of the Ising model are among the software tools and prototypes developed by the ADS-VSE team.

Student and the ADSRC-VSE Labs: There is a large number of students who are supported through research training, software development, and teaching-oriented training. The ADS-VSE faculty have worked with both undergraduate students and graduate students who are now experienced with computer graphics programming, Virtual Reality concepts, capable of using the SGI (Silicon Graphics, Inc.) workstation environment, and have gained computer networking experiences through constant exposure to these tools. Currently, R. Giroux (FAMU) and J. Brown (FAMU) are ADSRC students pursuing the Masters degree, and W. Ding (FAMU) completed the Masters degree in Sofware Engineering Sciences (FAMU) in 1998. Several past students are currently employed by major companies. For example, students were placed at Lockheed-Martin (Ms. K. Stephens - FAMU), Abbott (D. Howard - FAMU), IBM (K. Carson - FAMU). The spoken language, with JAVA/Web-based graphics and visualization subproject has drawn several interested students in the ADSRC project.

Impact and Visibility: The ADSRC VSE lab have been used for organized workshops, e.g., the FAMU VSE lab has been used by the College of Pharmacy (FAMU) for a 3-day workshop on molecular modeling, which was offered to their graduate and post-doctoral students. The ADSRC-FAMU VSE lab is currently used to support a departmental course in computer graphics, a spinoff from the ADSRC project. The VSE lab at UHD includes a Virtual Environment (VE), including a camera and large-screen display untis, and is used for various virtual reality research activities. The VSE lab at UCF supports various Masters and Doctoral student thesis/projects related to on graphics and visualization.

Further, ADSRC students (Michael Arradondo and Charles Birmingham) participated in this training. Mike Arrandondo collaborated with a research team at LANIA, Xalapa, Mexico on a graphics/visualization and vision project, as an extension of the ADSRC VSE leveraging and collaborative effort. This effort is linked with a NASA-supported research laboratory at Univ. of Houston, led by Prof. Bowen.

The various subtasks, discussed above, have been summarized in section (1) of this report. The summaries contain subsections for activities which are planned for the coming year, and beyond.

The Ising-based simulation research, including the use of distributed and parallel computing techniques, will be continued by developing the underlying mathematics and physics to accommodate assumptions necessary for more realistic simulation. A distributed display of various phenomena, e.g., fire and smoke spread, across multiple computer screens will help elucidate their behaviors observed from multiple viewpoints in simulated situations. This work will integrate findings from the heat transfer subtask.

The effort on heat transfer, will continue with the correlation testing for a wider range oof operating variables such a liquids with widely differing physico-thermal properties (such as cayogenics), heat flux, and pressure.

We will continue to design and develop an advanced software development environment for building spoken language dialogue systems. The development environment will be a visual programming environment, designed to allow a developer to use a drag-and-drop interface for creating a distributed speech application.

Additionally, the ADS-VSE team will explore and extend collaborative work which will integrate AI (fuzziness), Vision (image capture for software control and remote control of terrain vehicles), Distributed Graphics, and Software Engineering techniques which are imperative for developing large systems. Lastly, innovative steps will be taken to increase student teaching, training, research, and placement in graduate programs. On this last point, we would seek relationship with institutions like the Naval Postgraduate School (NPS), Monterey, CA, where much work is undergoing on software requirements engineering for large systems and the University of Pennslyvania, where advanced software development environments research efforts will provide opportunities to our students.

Research Objectives and Significance:The main objective for this effort is to design, develop and implement efficient physics-based mathematical models for the computer simulation of fire, smoke, and other phenomenological elements in distributed simulation environments. This theme has been the primary focus of our work during the past year. It is also desirable to create a computer simulation model of the virtual battlefield with greater fidelity and real world characteristics. This feature will allow for realistic simulations of battlefield operations. An ongoing objective is to build a joint research program in the Mathematics and Computer Sciences Departments within the ADSRC Consortium that will focus on useful efficient modeling techniques for systems of particles using physics-based principles. These models and simulations will be of wide relevance to government and industry organizations.

Army Relevance:The incorporation of real-time obscurants into computer simulations will make the virtual battlefield much more realistic. This will improve the training of solders and facilitate their acquisition of military skills which will enhance their readiness for battle. The interactions between smoke, rain, and other elements are unpredictable during war situations. Simulation models that represent battlefield situations help to improve the understanding of such phenomenological occurrences and their impact on the soldier.

Methodology:Heretofore, the two-dimensional Ising model, based on Monte Carlo simulation methods and matrix analysis, has been the primary tool for analysis with respect to the parameters that characterize fire and smoke as a battlefield obscurant. We considered the Ising model as a mathematical model which represents a system of particles, the constituent parts of fire and smoke, for the simulation. Hence, the effects of temperature, magnetization, and force field on the particles as a result of the spinning effect (random motions) of the particles were evaluated mathematically. This allowed an accurate modeling of the various orientations or configurations of the particles, and eventually for simulating the behavior of fire and smoke. A search of the technical literature on fire and smoke has revealed a wealth of research activities using various modeling approaches. By far the most prolific technique involves the use of computational fluid dynamics (CFD) techniques to study fire behavior principally based upon the Navier-Stokes equation. During the current, we have begun to investigate CFD techniques and other physics and mathematical approaches to analyze fire, smoke and other phenomenological elements in our computer simulations.

A two-dimensional simulation of smoke based upon the Ising model has been implemented in C code in the SGI/MPI environment. This effort has also incorporated fire behaviors into the model. Also a more precise mathematical formulation of the simulation problem has been done. Several program segments in C programming language have been written and implemented to test aspects of the smoke model. For performance considerations, a parallel version of the code has been developed and is being tested.

The following tasks will be focused on in our future work.

 

1. Collect, analyze, and distribute specific information in the form of references on existing models of fire and smoke.

    

    

2. Develop mathematical mechanisms to simulate the physical behavior of smoke and fire in terms of temperature, external magnetic force, and motion of smoke particles based on the Ising model.

    

Research Objectives and Significance:The main objective for this effort is to design, develop and implement efficient computer simulation of fire propagation by extending the percolation model. The percolation approach has a well documented history in basic fire propagation behaviors through random mediums with the added effects of wind. An extension of this basic model to include topology, non-random mediums, moisture content and effects of both ground propagation and tree canopy propagation will allow for realistic simulations of battlefield fire propagation. These models and simulations will be of wide relevance to government and industry organizations.

Army Relevance:The incorporation of an extended percolation model for fire simulation will make the virtual battlefield much more realistic. This will improve the training of soldeirs and facilitate their acquisition of military skills which will enhance their readiness for battle. Simulation models that represent battlefield situations help to improve the understanding of phenomenological occurrences and their impact on the soldier.

Methodology:The implementation of a percolation model begins by superimposing a lattice of sites and bonds over a given finite region. The sites correspond to vertices and the bond to the edges of an undirected graph G(S,B). The percolation model can be based on either a site approach or a bond approach. We have chosen to use the terminology of the site approach because of it’s intuitive mapping into the terminology of fire spread.

In a fire spread model we start with a finite lattice, thus the percolation process is guaranteed to produce a finite graph. However, if we assume that the lattice can be unbounded, we are presented with the possibility of producing an infinitely large adjacency graph. If we view the system in terms of graph G = (V,E), where V is a set of vertices (sites) and E is a set of edges between the vertices of V, we can then state that if the products of the expectation of the existence of edge x, for all edges in E, is non-zero, the probability of the percolation process producing an infinite lattice L is non-zero.

As the wind blows a forest fire, the speed of its spread increases. This is to say, that not only is a fire more likely to spread in the same direction as the wind, it will also spread in that direction more rapidly. Given a model where discrete time steps are representative of real-time intervals, an alternative approach to modifying the probability of occupation is needed. To this end we have developed the approach of neighborhood skew. In this approach the speed of the wind determines the size of the neighborhood. The skewing process gives the neighborhood its directional configuration. The result of the variable neighborhood size, the skew of the neighborhood, and the orientation of the originating site relative to its neighborhood, reflects the direction in which we wish to direct the percolation.

The following tasks will be focused on in our future work.

 

1. We plan to continue with our experiments with other known fire

    

    

2. We plan to examine the shortcomings of the current model and plan to update the model to rectify the limitations of the current model.

    

To partition the lattice structure of the Ising Model for distributed/parallel computation in order to minimize the overall simulation time. This step is important as it leads to near-real time visualization of the virtual battlefield phenomena.

The Ising model is a powerful tool that facilitates the study of statistical mechanical systems which are often characterized by the existence of phase transitions. Phase transitions of statistical

mechanical systems include phenomena like the transition of a material from an anti- ferromagnetic state to a ferromagnetic state or the evaporation of water into steam. In order to model phase transitions of a statistical mechanical system, the energy levels or states of a large number of particles must be analyzed. The Ising model accomplishes this task in two ways. First, the particles are arranged in a lattice and assigned energy levels. Second, the interaction of the

particles is analyzed based up on the energy levels. By using two or three-dimensional lattices and studying the interactions of a cell with its neighboring cells, a model of a statistical mechanical system with phase transitions is possible. However, real-time simulations of large systems with large two dimensional or three dimensional lattices require significant processing power. Using a distributed computing environment with a Message Passing Interface (MPI) system can reduce the time required to process large lattices mathematically and graphically

The parallel version of the Ising algorithm is running in the MPI environment. The simulation is being mapped into a datafile for conversion into OpenGL format for visual representation and understanding of the inherent phenomenology.

An accurate understanding of flame propagation mechanism and the modeling of fire have many implications for public safety, design of industrial heat transfer equipment, defense applications, and academic interests. There are many important problems related to the general process of flame propagation along the interface between a gaseous oxidizer and a liquid or solid fuel. The problem covers a very large field, including processes such as flame spreading in fires, design and operation of gas-fired heat transfer equipment (combustors), and flame spreading over the surface of thermoplastic fuels during the ignition process in hybrid rocket motors. Over past four decades a vast amount of research work has been conducted to understand the mechanism of fire propagation. The objective of this work is to conduct an elaborate literature search and review the approaches that different investigators have undertaken to study and model this complicated process of fire propagation. In short, this work is a compendium of the field of fire and smokes modeling and serves a report on the current physics-based modeling in this important area of research.

Distributed simulation environments have played a significant role in an ever-increasing number of critical military and commercial applications. Applications ranging from fire safety training to virtual warfare benefit tremendously from this emerging technology. The virtual battlefield necessitates the inclusion of real-time obscurants such as fire and smoke into the computer simulation in complex ways. Since real time is an important factor in distributed simulation environments, an inclusion of physics-based model of fire should have fewer computations and provides a realistic approximation algorithm.

As mentioned above the main objective of this research work is to conduct a comprehensive survey and analysis of the existing physics-based models for fire behavior, and flame and smoke propagation. A thorough literature search has been undertaken. A critical review of the literature suggests that a true mathematical model for flame and fire propagation would require the solution of a system of coupled, multidimensional, elliptic, nonlinear partial differential equations that would include variable material properties, appropriate gas phase chemical kinetics, solid phase pyrolysis mechanisms, heat and mass transfer convective terms for multi-component systems, and heat transfer terms due to conduction, and radiation. The solution of this full problem, even after considerable simplifications, is extremely difficult. For this reason the analyses developed to date have treated the problem at different levels of complexity. Researchers have investigated fire and flame propagation from a variety of perspectives depending upon the areas of application and their research interests. At this point, we have broadly classified the approaches that different investigators have undertaken to study this complicated process and written a survey article describing these approaches. These approaches are:

 

1. dimensional analysis for empirical correlations and scaling applications in fire research,

    

    

1. extension of low temperature heat transfer correlations,

    

    

2. computer simulation fire models,

    

    

3. mathematical modeling of flame spread over the surface of combustible material, and

    

    

4. smoke modeling of pool fires.

    

This is a preliminary work to understand the behavior of fire and flame propagation. An initial survey article has been written on this work. This work has been accepted for presentation at 1998 Conference on Simulation Methods and Applications (CSMA 1998) to be held at Orlando, Fl, from November 1-3, 1998.

Planned Activities:

 

1. The results of this survey will be used to identify the key parameters which affect the flame propagation and how these parameters can be used to develop a real-time fire model in a distributed simulation environment.

    

    

2. A more comprehensive analysis of available research work on flame and smoke propagation in a variety of fires is planned.

    

A rigorous mathematical model, taking into consideration, important factors needs to be developed.

To enhance the Computational Geometry Tutorial (CGT) system in order to provide more advanced topics/concepts. The enhancement included the introduction of Virtual Reality (VR) modeling component for animation, quality rendering, and visualization of graphic objects.

The CGT system is being developed to support and offer a tutorial system on how to construct and visualize geometric objects for ADSRC students. The ultimate goal is to provide students with the capability to develop graphic objects for visual enhancement of virtual battlefield in war-game simulations.

The CGT tool implements fundamental operations: translation, rotation, and scaling which allow graphic object manipulation using matrix operations. Further, the system uses several low-level graphic constructs and their applications within a simulation environment such as VRML, Java applets, Bezier curves, B-Spline, NURBS, Voronoi Diagrams and Convex Hull.

Accomplishments:

The CGT system has been enhanced greatly and currently available on-line (web-based) access and usage for faculty and students. The summary of the results was published in a conference

proceedings at the ADMI98 Conference, held at Houston, TX, June 26-27,1998.

To develop the mathematical basis of the geometric functions for the CGT system.

The CGT system is a useful resource for distance learning and offers a tutorial environment for learning computation geometry concepts outside of classroom setting. Its continual development to benefit all ADSRC participants is important, as it minimizes the learning curve and offers a quicker transition to the development and implementation of more advanced research topics in DIS.

Mathematical concepts such as convex hull, B-spline, Bezier-curves, etc., were analyzed, simplified, and integrated into the CGT system to enhance the synthetic environment. The synthetic environment methods focused on the application of accurate visual and aural data in physical and behavioral models.

In addition to the new theoretical concepts introduced and currently being implemented, a paper was presented at the Workshop on Distributed Simulation, AI, and Virtual Environments, Mexico City, February 22-23, 1998. The collaboration with the researchers at LANIA was significant for our visibility.

Transport processes with phase change such as boiling (nucleate, film, and flow) are the most efficient heat transfer modes that are characterized by small temperature differences and high heat fluxes. Therefore, they are employed in most thermal energy conversion and transport systems, as well as in heating and cooling devices, e.g., the dissipation of high heat fluxes from VLSI microelectronic packing using nucleate boiling with air jet impingement. Boiling has been one of the most efficient methods for cooling nuclear reactors, space vehicles, and electronic and microelectronic devices. This research work will be helpful in designing efficient and reliable heat transfer equipment where nucleate boiling is an important heat transfer mechanism. The objective of this research work is to develop a physics-based model for calculating heat transfer rates in nucleate pool boiling heat transfer of pure liquids over a wide range of pressure and heat flux.

The mathematical model developed in this work for calculating heat transfer rates in nucleate pool boiling can possibly be used to design efficient heat transfer systems for the cooling of electronic devices such as supercomputers.

The mathematical model for heat transfer coefficient uses the bubble departure diameter and bubble emission frequency. In this work the equations for these two pertinent quantities are obtained by considering both static and dynamic forces on a growing vapor bubble. The static forces are surface tension and the buoyancy and the dynamic forces are liquid inertia, bubble inertia, and the drag forces. Theoretical equations are developed for all these forces considering underlying physics and thermodynamics. These equations were then eventually obtained in terms of physico-thermal properties of liquid and vapor and were used to develop equations for bubble departure diameters and bubble emission frequency. Magnitude of these forces were calculated for refrigerants, hydrocarbons, and distilled water for a wide range of heat flux and pressure. These calculations provide basic understanding of boiling phenomena, e.g., how the magnitude of these forces change heat transfer coefficients with heat flux and pressure.

The correlation for heat transfer coefficient has successfully predicted the data of various investigators conducted for the liquids and surfaces of differing properties over a wide range of heat flux and pressure. This research work has been accepted for 11th International Heat Transfer Conference held at Kyongju, Korea, from August 23-28, 1998 and published in its Proceedings.

The correlation will be tested for a wider range of operating variables such as liquids with widely differing physico-thermal properties (such as cayogenics), heat flux, and pressure.

The main objective for the third year of this project was to further our knowledge and to attack problems as related to software issues, in general, and specific applications in advanced distributed simulation in particular. Our foci are on: Knowledge Acquisition and Machine Learning; Modeling Tools for Systems Under Uncertainty; Decision Making Models in the Presence of Imprecise Information; Computer Security Model Based on Uncertain/Partial Information; Concurrency Control and further development of a prototype of an intelligent autonomous navigation system. We show how the following tools significantly impact the above problems: Fuzzy Logic, Fuzzy Neural Networks, Petri Nets, Markov Chains, The Dempster-Shafer Theory of Evidence, Relevance Filtering, Analytical Hierarchical Processing, Object Oriented Databases

Our efforts produce a major impact on how future software may be written. In Machine Learning we provide methods of refinement to update weights for neural networks and improve our knowledge of membership functions in fuzzy rules. In Systems under Uncertainty we reduce the number of states of the Markov Chain with the relevant Petri Nets and generalize the standard fuzzy systems. This provides models of Systems using fewer and more powerful rules. In models for decision-making and, in particular, for estimations of project completions, we replace the classical PERT approach with our BIFPET method and compare the advantages of the latter method. Our Computer Security Model that operates in an environment where the user characteristics are uncertain offers a flexible access control to users as a function of their perceived potential hostility. In Concurrency Control, information related to particular databases may only be partially available and fuzzy logic is applied. Using fixed and dynamic grids, the appropriate level of information to be sent or received is defined. In the Autonomous Navigation Prototype System, hierarchical fuzzy controllers were developed to alleviate an explosion in the number of rules. Fuzzy CLIPS and OpenGL were used to exhibit the model.

Combat conditions provide very uncertain environments and first approximations of neural nets and fuzzy systems may be too rough. Combining neural nets and fuzzy rules offer the advantage of a system with learning capabilities while preserving a human like type of reasoning. Our approach is novel in that Dempster-Shafer Theory of Evidence and Interval Computations are used to create such systems. When making decisions under uncertainty we develop methods to compare fuzzy sets or quantities arising from them. In order to analyze complex systems, Petri Nets have been used which typically lead to a combinatorial explosion of states. We address this by aggregating sets of states, which renders Petri Nets more applicable under combat conditions. Computer Security is a prevalent problem for which we offer a flexible approach to software accessibility for users whose characteristics are only partially known. When dealing with large databases, concurrency control techniques allow a large number of simultaneous users. Fuzzy Concurrency Control provides additional insight on how to control the concurrency problem. In HLA it is important to define the minimum rate to maintain an accurate and consistent view of the ‘virtual world’. Our fixed and dynamic grid approaches offer a method to define fidelity. Algorithms for efficient data distribution, filtering effectiveness and computational overhead are analyzed within this architecture. Hierarchical fuzzy controllers alleviate the combinatorial explosion in the number of rules generated by previous approaches for the autonomous navigation prototype system. Interval computation better models the real world and provides reliable solution bounds for many application problems.

Imprecise and conflicting data was input into classical neural nets and the effects were analyzed. Fuzzy rules were tuned via the Theory of Evidence and weights of neural nets were set to intervals. In this fashion, we studied the inadequacy of classical neural nets and classical fuzzy rules and remedies to these deficiencies were proposed. An application to spinal chord stimulation for chronic pain management was outlined. Fuzzy Petri Nets for analyzing uncertain and variable environments were developed. This provided a significant reduction in the number of transition states. This model was applied to a problem of policy selection where the cost of each decision was only partially known. Fuzzy expectations were developed in a general setting. Fuzzy rules were analyzed where membership functions did not have precisely defined values. The response of these systems to imprecise rules was modeled. An alternate approach to the PERT method for analyzing project completions was proposed. Comparisons of complex quantities arising from fuzzy sets were proposed and an application was made to reduce the marketing uncertainties in the development of new products. A strategy for software access involving uncertain quantities such as expected losses, user hostility, and allowable damage amounts was developed. The performance of protocols for multi user access for database has been shown to depend heavily on how transactions and subtransactions are formed. We study these protocols when not all of the information is available. Hierarchical fuzzy controllers reduce the number of rules in our expert system associated with autonomous navigation system.

This effort focuses on the training of neural networks, the tuning of fuzzy rules, and some applications to the medical field. Fuzzy rules and neural networks typically need refinements in order to work well. Our approaches provide some methods of refinements to update weights for neural networks and to improve our knowledge of membership functions in fuzzy rules. A potential application of neural nets to the stimulation of the spinal chord has been proposed.

Since combat conditions provide very uncertain environments our first approximations of the appropriate neural nets or fuzzy systems are very rough and refinements are definitely needed. One of the main advantages of using fuzzy rules as opposed to crisp rules is that fewer of the fuzzy rules are needed to describe the behavior of a system. The disadvantage of using fuzzy systems is that learning does not take place. Since neural nets form a known body of knowledge where learning can take place, in recent years researchers have combined neural nets together with fuzzy systems to obtain the convenience of human like rules with the learning capabilities of neural nets. We use the Dempster Shafer Theory of Evidence to obtain a novel approach to construct a fuzzy system with learning capabilities.

The Dempster – Shafer Theory of Evidence was used. Individual training sets were used as sources of evidence to shape the membership functions. Also the weights of neural nets were assumed to be interval valued and appropriate convergence algorithms were used. In these works we develop methods to refine sets of fuzzy rules in order to obtain accurate systems. The traditional approach is to combine fuzzy systems with neural nets. The neural net is used to refine the parameters defining the membership functions involved in the rules. Here a different approach is taken and we use the Dempster –Shafer Theory of Evidence to refine the parameters. Each element of the training set is used as a source of information to generate a corresponding mass on the set of rules instead of a probability or strength coefficient on single rules. We also suggest how the evidence for a rule obtained by the training set could be combined with a genetic algorithm approach to generate a fitness function and therefore to obtain reasonable values for membership functions.

Imprecise and conflicting data was input into classical neural nets and the effects were analyzed. Fuzzy rules were tuned via the Theory of Evidence and weights of neural nets were set to intervals. In this fashion, we studied the inadequacy of classical neural nets and classical fuzzy rules and remedies to these deficiencies were proposed. Based on our D-S Theory approach a fitness function for possible Genetic Algorithms was proposed. An application to spinal chord stimulation for chronic pain management was outlined.

We plan to develop intelligent control systems for automated moving of objects with or without obstacles using fuzzy neural networks. Also we plan to develop a prototype of machine learning systems with fuzzy rules as input patterns as well as we plan to apply our tools for further study of the spinal chord stimulation process. In addition, we plan to look further into the relationship between the Dempster – Shafer Theory of Evidence and Genetic Algorithms as briefly mentioned above.

Stochastic Petri Nets have been used in the context of Stochastic modeling since the 1970’s. However a difficulty often encountered is a proliferation in the number of states in the associated Markov Chain. In the present works, we propose a solution to this proliferation problem by introducing the concept of a fuzzy Stochastic Petri Net. A related problem is to study Markov Chains where states are fuzzy subsets of some finite state space. An additional pursuit is: Given a set of fuzzy rules where membership functions are not totally determined and a fuzzy input where the membership functions are known, then one seeks to predict the output given by the system. This obviously generalizes traditional fuzzy systems.

Petri net techniques can be adapted for quantitative systems analysis by the introduction of temporal specifications by associating a firing delay with transitions. Since a firing delay is probabilistic in nature, the introduction of this component into the basic Petri Net leads to the development of a Stochastic Petri Net. The latter have been used mainly in the context of systems performance evaluation however Stochastic Petri Nets are a powerful and versatile tool for stochastic modeling in general. The main difficulty is the proliferation of states as mentioned earlier and we address this problem here.

We begin by considering here a Markov Chain whose states are fuzzy sets defined by some finite state space X. We consider Chains to have only a finite number of fuzzy states. Fuzzy transition probabilities arise naturally when the transition from one state to another is described by such mathematically imprecise phrases as ‘very likely’, ‘likely’, ‘unlikely’, etc. We have looked at procedures for decision making under this type of uncertainty. We describe a method to obtain crisp transition probabilities when the states are finite fuzzy sets. Related to these investigations, we have developed a method of reducing the number of states involved in a Stochastic Petri Net by lumping together different states into fuzzy aggregate states. This structure embodies partial information about a collection of crisp subnets. In particular, we have shown how to compute the stationary probabilities of reaching specific output nodes starting from certain input nodes. These stationary probabilities are fuzzy.

Fuzzy Petri Nets for analyzing uncertain and variable environments were developed. This provided a significant reduction in the number of transition states. This model was applied to a problem of policy selection where the cost of each decision was only partially known. Fuzzy expectations were developed in a general setting. Fuzzy rules were analyzed where membership functions did not have precisely defined values. The response of these systems to imprecise rules was modeled.

We will analyze how to fire all of the rules not just the dominant ones. A relatively recent concept that generalizes fuzzy sets of type II is that of ‘evidence set’. A natural problem is to investigate how to fire a set of rules described by such evidence sets. In particular one would like to define an appropriate concept of strength of a rule in such a setting. A starting point might be to obtain generalizations of Jain’s comparison algorithm, which compares fuzzy quantities to the setting of evidence sets. In addition, we would like to investigate how to defuzzify evidence set. A problem not considered until now is how to use training set to defuzzify better sets of type II. In the present work we have already defined the concept of conditional expectation applied to a fuzzy domain. We foresee many potential applications for this concept, particularly in the area of management and finance for which crisp data are not available.

In many situations, the total information necessary to reach a decision is not available. The underlying probability distribution may only be approximately known, as is the same for the payoff resulting from taking a specific course of action. In other situations, the relative order of importance of different factors contributing to selecting a course of action is only roughly known. The main purpose of these works is to develop tools to make an intelligent decision under such conditions. An additional consideration is that throughout the years the Beta Distribution was used to model variable activity times in the Program Evaluation and Review Technique (PERT). We present here two variations of a fuzzy probability model for project management. We use the concept of Belief in Fuzzy Probability Estimations of Time (BIFET) to analyze human judgment rather than go by stochastic assumptions to determine project completion times.

We have obtained several methods to compare fuzzy sets or quantities arising from fuzzy sets. Such comparisons are important, as they constitute tools to make decisions under uncertainty. Sometimes uncertainty arises from not precisely knowing probability distributions and/or payoffs associated with decisions. Sometimes uncertainty arises from experts being unwilling to put precise numbers indicating the relative importance of attributes in pair wise comparisons. In such cases one may want to take a consensus. In addition researchers have leveled some criticism at the PERT method. In practice the project manager must use three alternative estimates or three alternative Beta distributions depending on the skewness of the activity times. It has been suggested that skewness of the data can impact the justification for accepting the Beta distribution to determine mean activity times. We address the above-cited problems.

Methodology:

The classical Jain’s approach to comparing fuzzy quantities deals with sets of finite support only. We begin by considering sets with bounded support and then extend this to sets with unbounded support that vanish at infinity. An alternate approach which ‘cuts off’ sets with unbounded support is proposed. This latter method falls somewhere in between the classical defuzzification and the maximizing set approach. We generalize the Analytical Hierarchical Process (AHP) method to the fuzzy setting to be able to make a decision based on a consensus of experts who are not certain about making a pair wise comparison between alternate decisions. Moreover we handle the case where the experts have different capabilities. This latter case involves comparing quantities that generalize fuzzy sets of type II. Also a comparison is made between the BIFPET and the PERT approaches for round block production of a machine project. One of the many findings is that with BIFPET the activity supervisor may be fairly confident in the optimistic completion time and if the project manager has a strong belief that this confidence is warranted, the expected completion time should be adjusted downward toward the optimistic time.

An alternate approach to the PERT method for analyzing project completions was proposed. Comparisons of complex quantities arising from fuzzy sets were proposed and an application was made to reduce the marketing uncertainties in the development of new products.

Planned Activities:

We plan to further study the consensus of experts with different capabilities when they are unwilling to precisely assign scores while making pairwise comparisons between alternate decisions. In particular the objects arising from such considerations seem to be of mathematical interest as they generalize fuzzy sets of type II. We plan to demonstrate the usefulness of the BIFPET methodology in cost efficient crashing of project activities. Such an extension is important for insuring adequate assessment of the daily need for specific resources in order to crash activities on the critical path. Also we plan to look at applications, somewhat similar in spirit to our work on reducing the uncertainty in new product development.

The main objective of this project is to develop a model to autonomously navigate a vehicle by using hierarchical fuzzy controllers. The control rules of the fuzzy controller was formulated by applying both a general autonomous navigation scheme that mimics how humans behave in maneuvering a vehicle and hierarchical structure of rules. Our scheme implements fuzzy if-then rules that characterize humans' behaviors to autonomously control a vehicle. The hierarchy of fuzzy controllers was utilized to avoid combinatorial explosion in the number of rules occurring while increasing the number of sub-conditions in the antecedent part of each rule. These fuzzy controllers can be extended with minor modifications for any type of vehicles or planes, and any environment or terrain. Computer simulations were performed on a vehicle with different road conditions using FuzzyCLIPS and OpenGL to exhibit our model.

The results of this project provide an alternative way to navigate a vehicle in distributed simulations. This result can also be extended to model any plan actions with route planing simulations. Since fuzzy rules are obtained from experts in the form of linguistic information, our model can be very flexible and reliable.

Fuzzy controller is a very useful and efficient approach for utilizing expert knowledge in if-then form. It provides an algorithm that can convert the linguistic control strategy based on expert knowledge into automatic control strategy. It shows in previous works that the fuzzy controller yields results superior than those obtained by conventional control algorithms. In particular, the methodology of the fuzzy controller appears very useful when the processes are too complex for analysis by conventional quantitative techniques or when the available sources of information are interpreted quantitatively, imprecisely, or uncertainly. The dynamic behavior of a fuzzy system is usually characterized by a set of linguistic rules based on expert knowledge. The expert knowledge is usually of the form

IF (a set of conditions are satisfied) THEN (a set of consequence can be inferred).

For example,

IF (Speed is High) AND (Braking_Distance is Short) THEN (Brake_Applied is Heavy).

This is very useful in designing not only control systems with linguistic dynamics also control systems with complex dynamical behaviors. The hierarchical fuzzy controllers can be developed to alleviate a combinatorial explosion in the number of rules occurring while increasing the number of sub-conditions in the antecedent part of each rule. Because the number of rules can grow exponentially

large with respectively to the number of variables and number of linguistic terms used by each variable. For example, in Figure 1 a control system with 4 inputs and each input has m number of linguistic terms. A single controller with 4 inputs yields m⁴ rules, but a hierarchical controllers with 2 inputs at each level yields only m² + m² + m² rules. The following figure is an example of a rule structure that was implemented.

It is natural to implement FuzzyCLIPS to code all the fuzzy rules that we have derived since FuzzyCLIPS is a programming language that was developed especially for linguistic rule based. We developed two graphical simulation scenarios, highway environment and war zone scenario. OpenGL was utilized to generate graphics such as terrain, vehicles, tanks, airplanes, and obstacles. OpenGL is a set of graphic libraries that is a flexible interface to the graphic hardware of the Silicone Graphic machine. The libraries support over a hundred routines that setup the light source and every object to be displayed. Both FuzzyCLIPS and OpenGL code were organized to construct simulations. These simulations were run on a SGI machine. Each simulation is about 2-4 minute long.

We have developed a prototype for autonomous control systems by using hierarchical fuzzy controllers in navigating vehicles and route planning. The control rules of the fuzzy controller was formulated by applying both a general autonomous navigation scheme that mimics how humans behave in maneuvering a vehicle and hierarchical structure of rules. Computer simulations were performed on a vehicle with different road conditions using FuzzyCLIPS and OpenGL to exhibit our model.

Advances in computers and telecommunication have greatly expanded user requirements, applications and the tools available to all users. Users have become more and more dependent on the services provided by networked systems where data, computer programs, and highly sensitive information are kept in (geographically) dispersed systems and exchanged over telecommunication facilities. Distributed processing systems have emerged to provide the means through which networked systems cooperate to process users tasks in seamless and efficient fashion using the communication and computing resources provided by the underlying network structure. Such systems are becoming common in various areas such as banking (e.g. Electronic Fund Transfer system), trading (Electronic Data Interchange), and computer simulation.

Distributed systems provide tremendous new opportunities and benefits to their users but also raise new challenges. Because of the many different components, functions, resources, and users and the tight coupling between the cooperating systems, security in distributed systems is more difficult to achieve than in regular computer networks. The framework for security in distributed systems is the Open Systems Interconnection (OSI) Security Architecture. The OSI Security Architecture, however, is not an implementable standard; it is just a framework that defines security concepts and services that are used in the open-system security field.

The main objective of this work is to present an access control security model that determines whether a user is permitted to access and perform particular operations on particular data sets. Given the level of hostility of a user in a distributed system, and the sensitivity level of the data effected by the requested service, the local host/security guard is called upon to evaluate whether such a request can be safely granted. In general, information such as the hostility level of a remote user and the sensitivity level of a particular data set are either uncertain or only partially known by the local host, or both. Using fuzzy sets to represent uncertain/incomplete information, we present a framework that allows a local host to determine access permission based on such information.

This work may have applications for security issues on distributed computer networks where the systems allow general access for trustworthy users.

We developed a security model with a user x_i from a defined set of users X = {x₁, x₂, ..., x_m} that have a desire to perform a particular operation o_j on data set d_k, where o_j is an operation from the predefined set of operations O = {o₁, o₂, ..., o_l}, and d_k is a data set from D = {d₁, d₂, ..., d_n} that is maintained by a given organization. In this framework, their level of trustworthiness or hostility characterizes users. Hostility level can be determined based on several factors that can be deduced from information that is available about the user, such as the user's clearance level and the security services provided by the remote host. In addition, we define t as the amount of damage that the organization can tolerate from having a user perform operation o_j on data set d_k.

If PH_i is the probability of user x_i being hostile, t is the amount of damage that an organization can tolerate, and EW_jk is the worst loss expected from performing operation o_j on data d_k, then the expected loss E_ijk can be computed to determine whether user x_i should be allowed to perform operation o_j on data d_k. Obviously, the degree of the expected loss varies with the value of user hostility and is always less than or equals to the worst expected loss.

In general, information such as expected losses, user hostility, and allowable damage amount are very difficult to assess precisely in numerical terms. So it is natural to express them in the form of fuzzy sets. In linguistic terms, a user can be defined as very hostile, somewhat hostile, or not hostile, and amount of damage can be expressed as very high, low, very low. In fuzzy expressions, somewhat hostile, for instance, can be expressed as:

where the supports (i.e. .2 and .1) are the probabilities of the user being hostile and the .9 and .8 are the membership values. Damage amount can also be expressed in a similar fashion, with the supports being expressed in dollar units.

We establish a procedure to determine whether a user x_i should be allowed to perform the operation o_j on data d_k as follows:

1. Find expected loss E_ijk by evaluating EW_jk ´ PH_i.

2. Apply Jain's method of comparison to compare E_ijk with t by constructing a maximizing set M from fuzzy sets E_ijk and t .

3. Compute E_ijk Ù M and t Ù M then make a comparison between the sets. If the greatest membership value of E_ijk Ù M is greater than the greatest membership value of t Ù M, then the user x_i is not allowed to perform the operation o_j on the data d_k. Since the expected loss form the user x_i to perform the operation o_j on the data d_k is larger than the allowable total loss of damage. On the other hand, if the greatest membership value of is less than the greatest membership value of t Ù M, then the user x is allowed to perform the operation o_j on the data d_k.

We develop a framework for security model using fuzzy sets to represent uncertain/incomplete information. The model allows a local host to determine access permission based on whether a user is permitted to access and perform particular operations on particular data sets.

Object-oriented database systems (OODBS’s) deal with complex data types hence involve long transactions. Long running transactions often imply the use of a large number of resources that are inaccessible to other incoming transactions. The main objective of concurrency control mechanisms is to ensure the serializability of concurrent transactions. In addition, a great deal of research has been done in past several years to develop new algorithms that provides more efficient resource access to concurrent transactions. Our study has also shown that information related to the particular database, application domain and the transactions being executed can be used in the process of identifying the groups. In many situations, however, such information is only partially available or imprecise. This research applies the fuzzy set theory to a concurrency control mechanism for the cases where needed data is not fully available.

This work can be applied to some of HLA/RTI services such as Ownership Management Services.

Methodology:

A concurrency control mechanism called Group Protocol (GP) for Object-Oriented Database Systems was previously developed, which is a combination of Two-Phase Locking (2PL) and Serialization Graph Test (SGT) techniques. The Group protocol has the potential of achieving more concurrency at different levels of granularity by subdividing each transaction into a set of subtransactions each consisting of a group of operations. There are two types of groups, the Update Group (UG) and the Read Group(RG), which are different because of the type of nodes they contain and the order in which they release their locks. Our performance evaluation results have shown that the performance of this protocol depends to a great extent on how the groups are formed. In this research we develop a methodology based on the fuzzy set theory that allows us to optimize the performance of Group Protocol based on the available information. The idea basically is as follows: There is a threshold for when we let the transactions change a UG to an RG (to allow more parallelism hence, improve the performance). Each transaction is associated with some information, I1, I2, ....,Ik . The idea here is to compare the fuzzy information values I1Ù ....