Date of final exam: 21/12/2000

E-mail: lpollini@dsea.unipi.it

Tutor: Prof.  M. Innocenti, Università di Pisa

  ___________________________________________________________________________________________________________

Real-Time Distributed Simulation of Dynamic Systems   ___________________________________________________________________________________________________________

Advisor:

Prof.  M. Innocenti, Università di Pisa

Summary of the thesis

The simulation of complex systems which include manual control is always difficult, and, more important, its validation, that is the process of assessing reliability of simulation results, is often time consuming. One approach to the problem could be a hybrid simulation environment where parts of the system consist of actual hardware, and the other components are simulated by software. Due to computer processing power limitations, it is often unreasonable to use only one central processing unit to simulate software components, to acquire input data, and to generate synthetic visions of the outside world for example. The system components could be distributed into an array of computers, these must be interconnected by a computer network the same way they would be in the pre-distribution system, obtaining a low-cost hardware solution for distribution of simulation workload among various workstations to reach high computational power. Every real-world component "runs" with the same timebase, that is, every hardware apparatus or human operator sees the time running at exactly the same frequency. This is not true for simulated systems. In this thesis, a simulated system that runs synchronized with the real world timebase will be defined to be running in real-time.

The thesis, after a review of the state of the art and commercial products in the field of distributed simulation, introduces briefly the digital simulation of dynamic systems and addresses some issues in their parallel simulation. Particular attention is given to the problems induced by the integration algorithms. The generic variable step and fixed step classes of algorithms are examined to highlight their respective advantages and disadvantages within parallel simulation, and the choice of fixed step technique is motivated. Then, the hardware in the loop and man in the loop simulation paradigms are introduced with particular attention to their application in a distributed simulation environment. The thesis then presents an analysis of possible techniques for simulator decomposition and gives some heuristic criteria for choosing the best decomposition depending on the single scenario requirements. Then analyzes possible network communications strategies, and a classification of possible communication atoms. Furthermore, the network distribution induced problems, and simulator components synchronization are addressed together with an analysis of the real-time synchronization problem and the design of possible synchronization schemes.

Various possible communication protocol alternatives are analyzed for the implementation of the distributed simulation communication terminals: CORBA, FireWire and TCP/IP features, that in fact cover most of the communication typologies, are described, and then the choice of TCP/IP is motivated. The technical problems due to the TCP/IP choice are highlighted and, in particular, the risk of deadlock is introduced. The deadlock problem is solved via a pre-simulation choice of order of activation of all communication procedures. An automated technique based on graph theory, the Priority Manager Algorithm, is designed to solve the deadlock problem.

Distributed simulation accuracy is then examined and the requirements necessary to achieve no accuracy loss, with respect to non-distributed simulation, are given as well as directions for the design of appropriate communication terminals that minimize and, in certain cases, null the accuracy loss. Then, the designed communication terminals are used to field test distributed simulation, and simulation results are given that show the small overhead introduced by distribution and the near-ideal results that can be achieved.

Finally the requirements of synthetic environments are introduced from the standpoint of man in the loop (MIL) distributed simulation, and DynaWORLDS, a three-dimensional real-time visualization software, which has been initially designed by the author at the Department of Electrical Systems and Automation (DSEA), and now is growing up as a commercial product, is presented together with some examples of distributed simulators developed at DSEA during the three Ph.D. years. DynaWORLDS is the basis for all the MIL simulators developed at DSEA.
 

  _______________________________________