| Summary: | Thesis (S.M.)--Massachusetts Institute of Technology, Dept. of Mechanical Engineering, 2007. === Includes bibliographical references (p. 73-75). === New simulation modeling environments have been developed such that multiple models can be integrated into a single model. This conglomeration of model data allows designers to better understand the physical phenomenon being modeled. Models are integrated together by creating connections between their interface parameters, referred to as parameter mapping, that are either shared by common models or flow from the output of one model to the input of a second model. However, the process of integrating simulation models together is time consuming, and this development time can outweigh the benefit of the increased understanding. This thesis presents two algorithms that are designed to automatically generate and suggest these parameter mappings. The first algorithm attempts to identify previously built integration model templates that have a similar function. Model interfaces and integration models are represented by attributed graphs. Interface graph nodes represent interface parameters and arcs relate the input and output parameters, and integration models graph nodes represent interface graphs and arc represent parametric connections between interface graph nodes. === (cont.) A similarity based pattern matching algorithm initially compares interface graphs in two integration model graphs. If the interface graphs are found to match, the algorithm attempts to apply the template integration model's parameter mappings to the new integration model. The second algorithm compares model interface parameters directly. The algorithm uses similarity measures developed for the pattern matching algorithm to compare model parameters. Parameter pairs that are found to be very similar are processed using a set of model integration rules and logic and those pairs that fit these criteria are mapped together. These algorithms were both implemented in JAVA and integrated into the modeling environment DOME (Distributed Object-based Modeling Environment). A small set of simulation models were used to build both new and template integration models in DOME. Tests were conducted by recording the time required to build these integration models manually and using the two proposed algorithms. Integration times were generally ten times faster but some inconsistencies and mapping errors did occur. In general the results are very promising, but a wider variety of models should be used to test these two algorithms. === by Thomas Ligon. === S.M.
|
|---|
