' •. CASE BASED APPROACH FOR MESH SPECIFICATION IN ADAPTIVE FINITE ELEMENT AJ."'{ALYSIS Abid Ali Khan, Xu Yuan Ming and Shen Zhang Beijing University of Aeronautics and Astronuulics, Beijing 100083, China Abstract: This paper describes the application of an Artificial Intelligence paradigm called CBR (Case Based Reasoning) to the initial mesh selection in the adaptive finite element analysis system. It gives a brief overview of the system modules which perform the :malysis on the ·structures by accessing the main module of the system called Case Base where previously solved cases are stored for future use. The system matches the current problem with similar stored case. The mesh distribution in the form of numerical values is represented by the object-oriented technique, which can be easily adapted by the topology of new problem in the same structural domain. As building of Case Based Reasona, eveJ. in its simplest form involves a number of issues, such as case representation, case indP.xing, case matching and retrieval. All these issues are briefly discussed in this paper. It is believed that the use of developed knowledge representation scheme in the CBR systems w i:; b" .:1 "tep ahead for Case Based Re:1::;or:ing :: .: ::·.-. __ ;<v0ll'rr IL•0 i in the structural analysis domain. Keywords: Finite element analysis, Object oriented programming, Knowledge representation, Case based reasoning, Error estimation, Mesh specification. 1. INTRODUCTION The generating of fmite element model involves great experience by the experts in the related analysis domain. Such tasks can be time consuming and error prone if done manually. Many approaches have been proposed to generate fmite element model. A number of expert systems such as EXPERT (Ba!Juska and - Rank, 1987), IQFEM (Breitkopf and Kleiber, 1987), SACON (Bennett and Englemore, 1983), SESCON (Fjellheirn and Syversen, 1983), FEMOD (Chen and Hajela, 1988) and FEMES (Gong and Xu, 1995) have been developed over the period of time to resolve the problems involved in model generation by using conventional rule based reasoning techniques. Other approaches (Shephard and Finnigan, 1988; Finnigan, et al., 1989; Unruh and Anderson, 1992) involve the incorporation of Computer aided designing (CAD) techniques, Intelligent decision making and advanced data management system into fmite element analysis, so as to fonn the hybrid integrated intelligent structure analysis system. Although all these approaches have shown some success in analyzing certain type of structures but still there are a lot of issues that need to be addressed and resolved to formulate a more universal and highly automated approach in the fmite ck111eal ana;ysis One of the diffic;dt t"!sks for the FEM users is to set the spatial mesh over the geometrical domain. Specifying mesh densities that capture the behavior of local phenomena, can typically become a process of trial and error. Intelligent users of the FEM for numerical discretization have always endeavored to assess the . accuracy of their results before using them in practice. Unfortunately, the most obvious procedure of. doing this by study of refmement convergence is generally too expensive to be used frequently. This paper proposes an intelligent-based reasoning -. .285 ·-