Finite element analysis using component decomposition and knowledge-based control
A novel component decomposition finite element method is proposed that employs knowledge based analysis for solution control and object-oriented programming methods for implementation. Application of these methods have enabled the development of a Thermal Analysis Blackboard System (TABS) that exhibits many improvements over traditional finite element methods. This system has five distinct characteristics: (1) The analyzer is for a specific application domain (electronic module) and phenomenological analysis (FEA) domain. Imbedded knowledge of these domains enable the analysis process to be intelligently controlled. (2) The control of the analysis process occurs at a higher level of abstraction than that of finite element nodes and elements. Instead control is based on the major components of the application domain, such as chips, substrates, and package. (3) The finite element analysis is conducted separately for each system component using decomposition techniques. This allows analysis of individual components to be optimized based on each component's unique characteristics. (4) The numerical analyzer that solves the matrix equations was written using object oriented programming methods. (5) Data storage and analysis control has been accomplished by using a blackboard architecture.
The use of component decomposition allowed for small incompatibility to exist at the intercomponent boundaries. This introduces an additional approximation into the problem solution. However, test cases showed that the small incompatibility at the component interfaces have a relatively small effect on the accuracy of the analysis results in the critical regions of interest, with a maximum error of 7% observer in one test case. This decomposition method may not be appropriate if the critical region of interest is near to the intercomponent boundary or if high accuracy is required. However, this relative error may be acceptable if the analysis is part of the design process. In this case timely results, with an acceptable amount of error, are more important than high accuracy of results at the expense of timeliness. (Abstract shortened by UMI.)
0984: Computer science