RESEARCH > Dissertations

System for Pattern Recognition in Variant Design


The task of this thesis is to propose a solution and to realize a system for case recognition in variant design. Proposed system will be a support for design of power transformers. System is developed by using the case-based reasoning (CBR) methodology, where case represents previous design solution which is, in appropriate manner, retained in case base. With described system, designer, according to task description, can find the most similar previous design solution and can use it as template for solving new design problem. The main part of reasearch deals with examing two phases from so called CBR cycle: retrieving the most similar design and retaining new case in case base. Retaining of cases in case base is realized by recording the values of design parameters in relational database. Retrieving of most similar cases is based on ((nearest neighbour» method where is importance of parameters, by which we count similarity of observed cases and new design task, defined by weighting factors. Design are created by parametric three-dimensional CAD system. Retaining of cases for future use is realized in a way that we collect parameter values from parametric models of new design. These values are then recorded in case base. Solving of new design task is realized with CAD system and is not in scope of this thesis. Proposed system for case recognition is implemented in CAD system Pro/ENGINEER®. Actions of retrieving the most similar cases and recording the new cases in relational database are made in JavaTM program language and J-Link® libraries. Relational database is developed with SQL Server® database.



Damir Salopek




MSc thesis