INCORPORATION OF PLAIN STRAIN AND AXISYMMETRIC ELEMENTS INTO FIXED GRID FINITE ELEMENT ANALYSIS F.S. Maan1, O.M.

Querin1* and D.C. Barton1

1

School of Mechanical Engineering, University of Leeds, Leeds, United Kingdom

ABSTRACT This paper shows the incorporation of plain strain and axisymmetric elements into the Fixed Grid Finite Element Analysis (FGFEA) method. An efficiency and applicability comparison between the FG and conventional FE elements is gained through the analysis of the MBB beam and rod with fillet examples. These plain strain and axisymmetric examples are analysed at increasingly refined mesh densities to determine the deflection and von Mises stress values and stress concentration factors, respectively. The results establish a correlation between FGFEA and FEA, such that the developed FG elements are validated and the extension of FGFEA to plain strain and axisymmetric analysis is substantiated. Keywords: axisymmetric, fea, fixed grid finite element analysis, plain strain. 1 INTRODUCTION The FGFEA method1 overcomes the problems of mesh adaptation and regeneration which occur in conventional FEA as a fixed-grid mesh is used, which is independent of the mediums boundary. The discretisation process consists of superimposing a fixed rectangular grid of regular sized elements over the domain of the medium. The elements are then classified into three distinct types, by considering their location within the domain of the discretised medium. The three elements types are defined as: Inside (I), Outside (O) and Boundary (B) elements. As the grid is independent of the medium, any geometrical changes do not cause element degradation, therefore averting mesh adaptation or regeneration. This method has been successfully implemented with Genetic Algorithm2 (GA) and Evolutionary Structural Optimisation3 (ESO) design methods. However, the only element used was the two dimensional four-node rectangular plane stress element4, with eight degrees of freedom (DOF). In this paper a brief account of the FGFEA method is provided and the extension of FGFEA to plain strain and axisymmetric analysis is discussed. Two example structures are analysed and the results are validated by comparison with conventional FEA. 2 FIXED GRID FINITE ELEMENT ANALYSIS The FGFEA method was developed by Garcia1 and was formed with the plain stress element. The standard FEA stiffness equation is used to generate the fixed-grid equations and is stated in (1).

[K ] = [B] [C ][B]d
e T e

(1)

The strain-displacement matrix [B] contains the first order partial differentials of the shape functions Ni and the relationship is shown in (2).
x 0 N1 0 N 2 0 N 3 0 N 4 0 [B] = 0 y 0 N1 0 N 2 0 N 3 0 N 4 y x

(2)

The element displacement field is a bilinear function of the nodal values. The exact form of the shape functions is shown in (3) and they are a function of the local coordinates (x,y), the half-lengths (a) and widths (b) of the element.
Ni =

(a x )(b y )
4ab

(3)

The constitutive matrix [C] contains the material properties and is used in the stress-strain relationship (4).

{} = [C]{}

1 [C] = E 2 1 1 0 0

0 0 (1 - )

(4)

This procedure is basically conventional FEA, but to obtain the FG equations the [B] and [C] matrices are substituted into (1) and expanded which produces a form of the element stiffness equation which is composed of k terms (5)

k = ec

N i N j e d x y

(5)

The material properties are constant over the element domain e, such that c can be taken outside of the integral (6)
k = ck * , where k * = e N i N j e d x y

(6)

This definition permits the Inside I and Outside O element stiffness equations to be defined. The material properties of the structure are used for the I constitutive matrix and the properties of the O constitutive matrix are set to an arbitrary small value , as there is no actual structure in this region. The I and O element stiffness equations are shown in (7) and (8), where is typically 110-6.

[K ] = [B] [C ] [B]d
e T O e O

[K ] = [B] [C ] [B]d
e T I e I
e

(7)
I

= Ke

[ ]

(8)

As the B element is comprised of the I and O properties, which are constant in their respective domains of the element, the boundary element stiffness equation is defined by (9), where e1 and e2 are the I and O domains, respectively.
k = cI
e1

N i N j e N i N j e d 1 + cO e d 2 2 x x y y

(9)

An assumption is then made that the I and O shape functions are directly related to the shape function for the element domain, as a function of the area occupied in the element. This results in the element stiffness of the boundary being defined by (10)

[K ]
e

= K e I + (1 ) K e

[ ]

[ ]

= ( + (1 )) K e I , where =

[ ]

e1 e

(9)

In the computational application of the FG approximation, the only matrix that must be explicitly calculated is the stiffness matrix for I elements [Ke]I (7). As the mesh is regular and invariant, this calculation is only done once at the start of the analysis and [Ke]I is then reused as required.

3 EXTENSION OF FGFEA TO PLAIN STRAIN AND AXISYMMETRIC ANALYSIS The process of section 2 was applied to a 4 node element with 8 DOF, where the [B], [N] and [C] matrices were defined for plain strain and then axisymmetric elements. This yielded the element stiffness equations [Ke]I, [Ke]O and [Ke]B for each of the element types. In order to assess these elements two example structures have been analysed using ANSYS software: a model of the MBB beam with six holes for plain strain analysis, and the rod with a fillet example5 for axisymmetric analysis. The material properties used in both examples are those of steel: E = 210103 N/mm2, = 0.3 and the value of the FG O element is 110-6. The analyses were implemented at five and four increasingly refined mesh densities, for the plain strain and axisymmetric models to verify convergence of the results. Conventional FEA was also undertaken for both models to provide a comparison of results and ensure the validity of the FGFEA results. 3.2 Plain Strain Example

3.1.1 Plain Strain Model Information

The right-hand side model MBB beam structure used in the plain strain analysis is shown in Figure 1, where all dimensions are in mm.
20 40

40
y

10

120

Figure 1: Plain strain model dimensions. The model was constrained in the x direction along the left vertical edge and in the y direction at the bottom right-hand corner. A distributed load of 0.83 N/mm was applied along the top horizontal edge. The model was analysed at five mesh densities A-E, increasing from 48-3072 elements for FGFEA and 51-3177 for FEA.

3.1.2 Plain Strain Results

The nodal y deflection (Uy) and von Mises stress values were calculated for the bottom edge of the model at each mesh density. The FE and FGFEA results at mesh density C are shown in Figure 2.

Bottom Edge Nodal x Distance (mm) 0 0.0E+00 -5.0E-05 -1.0E-04 FEA FGFEA 10 20 30 40 50 60 70 80 90 100 110 120

1.4 1.3 1.2 1.1 von Mises Stress (MPa) 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 50 60 70 80 90 100 110 120 Bottom Edge Nodal x Distance (mm) FEA FGFEA

U y D eflection (mm)

-1.5E-04 -2.0E-04 -2.5E-04 -3.0E-04 -3.5E-04 -4.0E-04 -4.5E-04

Figure 2: Comparison graphs at mesh density C for displacement (left) and von Mises stress (right). The graphs show a very good correlation of the results at this intermediate mesh density, where there are 192 FGFEA and 195 FEA elements. The correlation occurs throughout the displacement graph, but reduces in the stress graph from a nodal x distance of 115mm. This is due to the constraint at the end node which causes a high stress area, and as the accuracy of FGFEA reduces in regions of high stress this was expected. These correlation trends were observed throughout the mesh densities as the results converged, except at mesh density D which has a large percentage error. This is evident in Figure 3 where the absolute average percentage error at each mesh density is shown.
7.0 6.5 6.0 Average Percentage Error (%) Average Percentage Error (%) 5.5 5.0 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 A B C Me sh Density Indicator D E 12.0 11.0 10.0 9.0 8.0 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 A B C Mesh De nsity Indicator D E

Figure 3: Graphs of average percentage difference for displacement (left) and von Mises stress (right). The error bars indicate the standard deviation of the average percentage error at each of the mesh densities for displacement, but are not present in the von Mises graph due to the nature of the standard graph, as shown in Figure 2. The displacement graphs results exhibit an oscillatory trend as the mesh density increases, but the high error associated with density D compromises this. The path of the von Mises error is also compromised by the error at this density, i.e. the convergence path of the error is it interrupted. However, a very low error is obtained at mesh density E in both graphs substantiating plain strain FGFEA. It seems likely that there is a problem with the FGFEA model at mesh density D rather than the results simply being poor, and this aspect requires further investigation. 3.2 Axisymmetric Example The rod with a fillet model is obtained from the ESDU data sheet5, and the specific geometric ratios selected from the sheet for this example are r/d = 0.2, and D/d = 2, where r is the radius of the fillet, d and D are the minimum and maximum diameters, respectively. The resulting stress concentration

factor is given as 1.779 and is located close to the fillet and minimum diameter junction, thus this is the value that the FGFEA and FEA results are expected to converge to.

3.2.1 Axisymmetric Model Information

The modelled section of the rod with a fillet structure used in the axisymmetric analysis is shown in Figure 4, where all dimensions are in mm.
y x

200 20 40 12 120 8

Figure 4: Axisymmetric model dimensions. The model was constrained in the y direction along the left edge and in the x direction along the top edge. A distributed load of 1 N/mm was applied along the right edge. The model was analysed at five mesh densities A-E, increasing from 80-5120 elements for FGFEA and 80-5119 for FEA.

3.2.2 Axisymmetric Results

The maximum stress concentration factor at each mesh density for FGEFEA and FEA is shown in Figure5.
1.7 1.6 Stress Concentration Factor 1.5 1.4 1.3 1.2 1.1 1.0 0.9 A B C D Me sh Density Indicator FEA FGFEA

Figure 5: Graph of maximum stress concentration factors There is a good correlation between the FGFEA and FEA results in the graph as both follow a similar convergence path. The maximum error between the results is -15.20% at mesh density A, which is acceptable due to the coarseness of the FG mesh. The error steadily reduces to a value of -8.28% at mesh density D, and from observation of the graph this error would continue to reduce as the mesh density increases. As the mesh density increases the location of the maximum factor tends to that stated in the ESDU data sheet, this also occurred with respect to the agreement between the two methods and is shown in Figure 6 at mesh density D.

Figure 6: Stress concentration factors and element distribution for FGFEA (left) and FEA (right). The ESDU data sheet5 stress concentration of 1.779 was not achieved at the maximum mesh density, resulting in a -11.02% and -18.38% difference with the FEA and FGFEA results, respectively. Further analyses are required at higher mesh densities to ensure that the ESDU value is obtained. 5 CONCLUSION In this paper the FGFEA method was extended to plain strain and axisymmetric analysis, by the development and incorporation of two new FG elements. The method was applied to a right-hand side model of the MBB beam, with six holes, for plain strain analysis (Figure 1), and to the rod with a fillet example (Figure 4) for axisymmetric analysis. The analyses were undertaken at increasing mesh densities for both examples. The Uy nodal deflection and von Mises stress values for the bottom horizontal edge of the MBB beam, and maximum stress concentration factors for the rod with a fillet structure, were calculated. The validity of these FGFEA results was established by comparison with those from conventional FEA. Convergence occurred in both models as the mesh density was increased and a general trend for the reduction of error was observed between the methods. However, the plain strain model at density D produced a higher than expected error, and this result did not fit the observed trend in both displacement and von Mises stress graphs (Figure 3). As mesh E produced a very good result with low error, it is thought that there was an error with the actual plain strain FG model at density D. The axisymmetric results did not converge to the expected ESDU stress concentration factor at the final mesh density. This is due to the mesh not being concentrated enough and resulted in differences of -11.02% and -18.38% for FEA and FGFEA, respectively. The maximum stress concentration factor occurred at the expected location (Figure 6) and the error between the two methods steadily reduced at each mesh density. Therefore, the results substantiate the extension of FGFEA to plain strain and axisymmetric analysis, but higher mesh density analyses will be investigated to rule out the possibility of anomalous results. REFERENCES [1] [2] [3] [4] [5] M.J. Garcia, Fixed Grid Finite Element Analysis in Structural Design and Optimisation, PhD., Dept. of Aeronautical Eng., University of Sydney, 1999. S.Y. Woon, O.M. Querin, G.P. Steven, Application of the Fixed-Grid Method to Step-Wise GA Shape Optimisation, 2nd ASMO UK / ISSMO conference on Engineering Design Optimisation, 10-11 July 2000, Swansea, Wales, UK, pp 265-272. M.J. Garcia, O.E. Ruiz, G.P. Steven, Engineering Design Using Evolutionary Structural Optimisation Based on Iso-Stress-Driven Smooth Geometry Removal, NAFEMS (Intl Assoc. Eng. Analysis Community) World Congress, 24-29 April 2001, Lake Como, Italy. R.D. Cook, D.S. Malkus, M.E. Plesha, R.J. Witt, Concepts and applications of Finite Element Analysis, 4th Ed. John Wiley & Sons, Inc, 2002. pp 238-247. ESDU, 890048 Elastic Stress Concentration Factors, Geometric Discontinuities in Rods and Tubes of Isotropic materials, ESDU, 1989, pp8.