BENCH, Yj, Modelling for Efficient FE Meshing thin the CAD Environment Pavao Marovic, Associated Professor, Goran Gusic, Researcher. Faculty of Civit Engineering, University of Split, HR-58000 SPLIT, Matice hrvatske 15, CROATIA, tel. (385) (058) 551- 475, fax. (385) (058) 524-162; SUMMARY This paper presents an approach to FEM (Finite Element Method) pre- and post- processing by means of the widely used CAD (Computer Aided Drawing) program AutoCAD. An object oriented paradigm in defining structure geometry enables a high level of data abstraction in the pre~ processing phase. Mesh generation, as well as generation of boundary conditions and loads which are defined over the goometrical entities, are performed using the, so-called, block mapping technique. A finite ‘element mesh created in this way is ‘well-conditioned and allows the use of different finite element types which is ‘one of the best features of mapping techniques. The wide open architecture of AutoCAD also enables fast definition of the input data, ‘The work presented has produced two programs, one inside AutoCAD which converts a geometry model into its ‘mumerical correspondent and the other one which generates the finite clement mesh outside the AutoCAD based on the preduced numerical model. Both programs —_ are incorporated in the FEME (Finite Element Modelling Extension) programme package, whose performance is illustrated by a few examples. 1. INTRODUCTION In recent years large physical problems have been solved by means of numerical methods. Huge amounts of data, concerned with these problems and their appropriate handling have arisen as objective FAST-INFO/ @® difficulties. The need for the visual- isation of solutions and simpler data input became a necessity. These facts give rise to two new phases in numerical modelling by computer pre-processing (definition of input datay and post-processing (visualisation of obtained results) uh ‘The Finite Element Method (FEM), is the most common method in the area of structure numerical simulation, appeared during the 1950's, wien the solutions of numerical simulations appeared as vast listings of the numerical data, presenting stresses and displacemeats in the discrete points of the model. The process of converting areal structure ‘model into numerical data has also resulted in data increase. This phenomenon resulted in many errors in the data input phase. More and more complicated geometry and increasing demands for complex structure description made it impossible to properly use te resalts produced by the FE model. Thus, the problem of data visualisation arose in the domain, first in the data input and later in the data output. For this purpose many graphical packages have been used. However, today the most widely used package is AutoCAD 2] According to the presented facts, this article briefly describes the FEME (Finite Element Modelling Extension) programme package (3, 4]. FEME is at the same time a pre- processor and post-processor for any finite element program and is based on the CAD program package AuloCAD. It has a built-in mesh ‘generator for pre-processing 2, MESH GENERATION TECHNIQUES The adoption of the geometry based definition of the problem and automatic production of a valid finite clement discretisation without user intervention is the major property of the mesh generators which represents apart of the reliable application ofthe finite element techaique. This results in the reduction of input data, mainly topology data of finite elements, and their co-ordinates. This also reduces the possiblity of producing errors. Most of the research in the area of mesh generation is focused on techniques such as 2D-quadtree [5]. 3D-vctree [6], trizngulation, paving (7) or on the -different mapping techniques. The mapping technique is used in this paper as well as in papers 1B, 4]. The main advantages of mapping techniques (block decomp- sition, transfinite mapping) are Boundary Sensitivity: mesh contours follow the contours of the boundary; Orientation Insensitivity: rotation or ‘translation of the given domain does not change the mesh topology; Element Versatility’ these techniques ccan be developed for different finite ‘element types. Geometrically based finite clement ‘mesh generators (5, 6, 8, 9] are, in comparison to the block mapping technique, at a higher level of automation, Despite this property these techniques are usually used for only one finite clement type ‘The quadtree mesh generator [5] is based on a spatial decomposition procedure of the 2D' domain into a set of disjointed squares. These squares are referred to as quadrants that are stored in a hierarchic tree. The root of the tree is given as a square that encloses the observed domain, Within the discretisation defined by the quadtree, the mesh topology is defined on a quadrant-by-quadrant basis, using the information stored in the March 1994 Page 23 *quadtree. The cctree technique [6] is very similar but only for 3D domains disjointing domain in octaedars. ‘The paving technique (7] is based on iteratively layering rows of elements ‘within the interior of a domain's ‘boundaries. As rows begin to overlap or coincide in the interior of the scometry, they are carefully connected together to produce valid topology of the finite element mesh. 3.0 BLOCK MAPPING TECHNIQUE 3.1 INTRODUCTION ‘The block mapping technique was introduced in 2D planar domains in 1971 {10}. The domain, according to this technique, is subdivided into blocks which are compatible in the same way in which the fiite elements are; then, follewing the subdivision parameters for each block, the blocks are mapped and subdivided into a valid finite element mesh, The tedious geometry of the finite clement model can be the source of errors and of the generation of non- valid. finite element meshes. This problem can be avoided by the use of the geometry based definition of the blocks by means of CAD modellers. In this way, errors are minimised even for manifold objects of arbitrary ‘geometric complexity. ‘The object oriented approach (OOP), hich has beea used quite often in programming, transfers the geometry definition of the domain into its x ‘mumerical representation in this nay. Geometrical entities, such as points, lines, ares, bezier curves are carriers of the properties of the finite ‘element model. For cach of these entities different properties can be assigned like kinematic boundary conditions, dynamic boundary conditions (loads), thickness, etc Four of these entities, each of them defining a block edge, have to be selected 0 create a block which is then mapped and subdivided to produce a finite element mesh. Data hiding, which is one of the ‘common OOP properties, :s used here to enable invisible correction between the mathematical medel and geometrical model, Although the mathematical model has no significance for the user it is hidden behind its geometrical representative im both pre-processing and post- processing phase. The user performs block discretisation by a CAD modeller without taking into account its topology and nodal co-ordinates which makes FEME a completely seometically based pre-processor 3.2 MATHEMATICAL MODEL ‘The mesh generation process starts afer the definition of the boundaries of each block and subdivision Parameters. Figure 1. Element in global and natural coordinate system with the point inside the element damain, Page 24 March 1994 BENCH af Each block is described by the ‘geometric entities which are carriers of the Bock boundaries. Mathematically, the blocks are given as, xX=ENx } @ ‘are nodal co-ordinates of 16 node isoparametric element, and x= {3} y are co-ordinates of each point within the clement domain, Ni are Lagrange shape functions. A graphical representation ofthis process for an 8- node element, is shown in Figure 1 4 x=t(6n) y=een) Once the block is mapped into a curvilinear co-ordinate system, the subdivision is performed in the following way bet aw i Sle laren a7 ae) m where ky qiy are parameters of a geometric series which enable non- uniform subdivision of the block domain, ic. the weighting factors which are used for weight sub- division Afier finding the curvilinear co- ordinates for each node of the finite clement mesh, co-ordinate mapping can easily be performed using eq, . ‘The aforementioned is performed using the MESHGEN programme which is incited in the FEME programme package. Accordingly, the MESHGEN programme performs finite clement mesh generation, evaluates co-ondinates of all nodes, performs all necessary numberings, and involves boundary conditions andFig. 2 Geometrical model b— : ( 1 2 3 a=——~zk Fig. 3 Superelements model. Fig. 4. Generated finite element mesh ‘material properties for all elements, In the first phase, superelement topology generation is performed with the help of subroutine BUILDTOP. In the second phase, in subroutine PROPAGATE, the subdivision of superelement on a given number of elements m x m is performed, for chosen parameters m and n, while simultaneously checking the connect- ion compatibility between the super- elements. After the transformation of the geometrical mode! of the domain into a correspondent mathematical ‘model, in subroutine MESH, the generation of finite element mesh is performed maintaining the boundary ‘conditions and material properties. In ‘the fourth phase, in subroutine REMDOUBLES, the double nodes are climinated at superelement contacts. Overall results of the MESHGEN programme are oblained in the databases with exiensions: *,CIF - database for communication with Programme AutoCAD; *ELM - database with all numerical data (standard alphanumerical input database) which can be modified in accordance with the needs of the used finite clement solver; and *.OUT - control database. Details of these procedures can be found in [3]. 4. CAD SOFTWARE AND ITS APPLICATION In the 1960-5, together with hardware development, a new group of ‘computer applications was born, CAD = Computer Aided Design. ‘These Programs became the tools for ‘geometry modelling both in 2D and 3D spaces. FEM was adopted in the 1950-s, ‘These ten years and the fact that design was separated from numerical simulation of the structure behaviour, has set apart FEM from CAD. Indeed, the analysis is not integrated in the design for good reasons, since, the design analysis is usually performed in different departments with different tools; thus, the gap detween geometry modelling and structure analysis is March 1994 Page 25‘wider than necessary [1] Model building must be based on a geometric model. Thus, it is time consuming 19 build a model consisting of nodes and elements, node by node and element by element ‘The CAD modeller can use geometric operations in both 2D and 3D spaces, such as union, subtraction and intersection. {A variety of functions, incorporated in the CAD modeller, also enables the development of the hierarchy of geometrical entities which define arbitrary structure geometry. Functions which have to be developed for a particular CAD software to connect CAD with FEM can be grouped, as follows: transfer; geometry model, built up by the CAD modeller can be directly used selecting the entities which define the ‘model, thus filtering necessary information about model geometry without picking up the rest ofthe data defined in the designing phase (dimensions, hatch, text, etc.), which are needed for a proper representation of the geometry model, trans- formation; gecmetry completed in this way, together with parameters for ‘mesh density and relevant data for FEM model such as material properties over the region is transformed info a valid numerical representation of the geometrical model. Associating the loads and the boundary conditions with geometrical entities, the checking and connection of the geometry model are performed in this phase, completion, formatting the data for use by FEM solvers. The first two phases are performed in AuloCAD [2], one of the most popular CAD software for PC compatible machines. FEME, which stands for Finite Element Modelling Extension, was developed for this purpose using AuloCAD's open architecture. Opening, the architecture of this CAD tool is performed by the included ‘AuloLISP interpreter [11, 12], which ‘enables the extensive use of all built- in CAD functions and the ‘manipulation of the graphical objects Page 26 March 1994 BENCH, 44), cexisting in the current drawings data base. AutoLISP is the part of Common LISP, a programming language developed in the 19508 by McCarthy for the sake of artificial intelligence rescarch which today represents one of the oldest programming languages. Its flexibility and the fact thet it was ‘wide open for user intervention have made possible its continual application. Graphical objects and lists of ‘graphical objects through AutoLISP Geometrical model by using encapsulation, data hiding and filtering are easily accessed and manipulated. AutoLISP, as the first AutoCAD interface language, is. suitable for engineering applications because: (a) itis the chosen language for research and development of artificial intelligence and expert systems; (b) it has an exceedingly simple syntax; (€) itis ideally suited for the unstructured interaction that characterises the design process; and (@) it excels at collecting the heterogeneous objects in various sized groups which is precisely the type of information CAD ‘Supereloment Finite clement mesh Fig. 5 Straight-curves cantilever: geometrical model, superelement ‘model and finite element mesh - undeformed and deformed. Fig. 6 The speed of the programme MESHGEN for different numbers of nodes and elements for 2 superelements.systems manipulate. ‘Two special AutoLISP data types are implemented to provide access to AutoCAD graphical entities: (a) entity name; (b) selection set. According to the OO paradigm these data types are manipulated only by functions that act upon them and their internal structure is irrelevant to the programmer. An entity name is actually a painter to the entity's database record and its vectors. A selection set is simply a collection of entity names, 5. EXAMPLES ‘What is the practical application of the aforementioned? On the basis of a known geometrical model, its contour lines are drawn withthe help of the AutoCAD programme. ‘The obtained sketch is then divided in accordance with material and geometrical properties on blocks - superelements. Sub element division parameters are then introduced for each siperelement. The obtained finite ‘elements take over all the properties of the superelement. So, the generated finite element mesh is obtaised visually ‘on screen in AutoCAD and alphanumerically in database “ELM, This database can be further used by an appropriate finite element solver which is connected with the FEME programme package. After performing the numerical calculation the obtained results - displacement and deformation data - can be relumed into the AutoCAD programme in order to be visualised. ‘The use of the FEME programme package as a pre- processor and also as a post-processor is illustrated by two examples, EXAMPLE No. 1 The first example chosen to illustrate the possibilities of the FEME programme package is a dam structure with surrounding soil. Figure 2 stows the geometrical model of the structure according to which the superelemeat model, shown in Figure 3, is evaluated. The result of the generation of finite element mesh is shown in Figure 4. The ‘obtained mesh is discretised with 175 elements and 598 nodes. The finite element mesh generated for this example {together with all data (node co-ordinates, clement topology, all numerical data concerning nodes and elements, material properties, etc.) is obtained in about 6 seconds. EXAMPLE No. 2 ‘The second example which is used to illustrate both pre- and post-processor possibilities of the FEME programme package is a straight-curved cantilever. Its geometry is shown in Figure 5 together with the superclement model and the obtained generated finite clement mesh, The deformed mesh obtained by one finite element solver for the linear stress and strain analysis [13] is evaluated in the same figure,6. CONCLUSIONS ‘This article presents the possibilities of the FEME (Finite Element Modelling Exiension) programme package, evaluated for pre- and post- processing of input and output data for any finite element solver by means ‘of CAD programme AuloCAD. It is worth mentioning thatthe preparation of input data by drawing instead of typing all alphanumerical data is such quicker and more precise. ‘The evaluated FEME programme package was tested on several examples considering the speed performances. Some results concerning speed of the meshing procedure obtained © for two superclements, by varying the number of nodes and elements are shown in Figure 6. That figure and practical experience show what improvements in working speed, cost reduction and accuracy can be obtained using the FEME programme package. ‘The FEME programme package is being tested now. It should be improved by fulfilling all the requirements roquested by appropriate NAFEMS benchmark tests [14] before ican be professionally used. 7. REFERENCES U1] G, Butlin, State-oFthe-art Pre- ‘and Post-processing, FAM technical paper, FEGS Ltd., England, UK., Doc. TP-87.10, 1987. [2] N. Johrson, AutoCAD: The Complete Reference, Second Edition, Osborne McGraw-Hill, New York, 1991 [3] G. Gusi), Preparation of input data and visualization of output data for finite clement solver using AuloCAD, Diploma Thesis, Faculty of Civil Engineering, University of Split, Split, 1993. (in Croatian) [4] P. Marovi} and G. Gusi3, Interfacing FEM with AutoCAD, Proc. 2nd Conf SCSE Civil Engineers in Rebuilding Croatia, Brijuni, Ed. J, Radi}, pp. 149-154, 1993. (in Croatian) Page 28 March 1994 BENCH ak [5] PL. Bachmann et al,, Robust Geometrically Based Automatic ‘Two-Dimensional Mesh Genzration, Int. J. for Num, Meth. in Engng., Vol. 24, pp. 1043-1078, 1991. 16] MS. Shepard and MK. Georges, Automatic Thee Dimensional Mesh Generator by the Finite Octree Technique, Int. J. for ‘Num, Meth, in Engng,, Vol. 32, pp. 709-749, 1991. [7] TD. Blacker and MB. Stephenson, Paving: A New Approa: to Automated ‘Quadrilateral Mesh Generation, Int J. for Num. Meth. in Engng., Vol. 32, pp. 811-847, 1991 [8] J.C. Cavendish, D.A. Field and WH. Frey, An Approach to Automatic Three-Dimensional Mesh Generation, Int. J. for Num. Meth, in Engng, Vol. 21, pp. 32 1985, [9] V.N. Kaliakin, A Simple Coordinate Determination Scheme For Two Dimensional Mesh Generation, Computers & Structures, Vol. 43, No. 3, pp. 505- 516, 1993, [0] O.C. Zienkiewiez and D.V. Phillips, An Automatic Mesh Generation Scheme for Plane and ‘Curved Surfaces by Isoparametric ‘Coordinates, Int. J for Nun. Meth. Engng, Vol. 3, pp. 519-528, 1971 [11] AutoLISP, Release 11, ‘Programmer's Reference, AUTODESK, 1992, [2] V. Liubi~, 1. Turk and T. Knific, AuloCAD and AutoLISP, FAGG & IKPIR, Ljubljana, 1991 (in Slovenian) [13], Hinton and D.RJ. Owen, Finite Element Programming, Academic Press, London, 1977. [14] B. Spooner, Benchmarks for Finite Element’ Pre-processors, NAFEMS Poblication Benchmark, October 1990.