Shape and Topology Optimization of Brackets using the Level Set Method by Michael R. Thomas An Engineering Project Submitted to the Graduate Faculty of Rensselaer Polytechnic Institute in Partial Fulfillment of the Requirements for the degree of MASTER OF ENGINEERING IN MECHANICAL ENGINEERING Approved: _________________________________________ Professor Ernesto Gutierrez-Miravete, Project Adviser Rensselaer Polytechnic Institute Hartford, Connecticut DECEMBER 2010 © Copyright 2010 by Michael R. Thomas All Rights Reserved ii CONTENTS LIST OF TABLES ............................................................................................................ iv LIST OF FIGURES ........................................................................................................... v LIST OF SYMBOLS ........................................................................................................ vi ABSTRACT ...................................................................................................................... 1 1. Introduction.................................................................................................................... 2 1.1 Background ........................................................................................................ 2 1.2 Shape and Topology Optimization .................................................................... 2 1.3 Level Set Method using Scilab Code ................................................................. 4 2. Theory ............................................................................................................................ 5 3. Methodology .................................................................................................................. 7 3.1 Level-Set Method ............................................................................................... 7 3.2 Sample Brackets ................................................................................................. 7 4. Results and Discussion .................................................................................................. 9 4.1 Example Brackets............................................................................................... 9 4.2 Establishment of Norms ................................................................................... 12 4.3 Bracket Comparison ......................................................................................... 12 4.4 Stress Analysis using COMSOL ...................................................................... 12 5. Conclusions.................................................................................................................. 17 5.1 Summary of Results ......................................................................................... 17 iii LIST OF TABLES Table 1 Summary of comparison study…………………………………………16 iv LIST OF FIGURES Figure 1 Typical sheet metal brackets.....…………………………………………..2 Figure 2 Moving the zero Level Set…....…………………………………………..3 Figure 3 Bracket boundary conditions ……………..…………….………………..8 Figure 4 Optimization results for example bracket 1…………………...……….....9 Figure 5 Optimization results for example bracket 2………………...……….......10 Figure 6 Optimization results for example bracket 3…………………...…...…....11 Figure 7 Optimization results for example bracket 4…………………...………...11 Figure 8 Optimization results for example bracket 5…………………...…...…....12 Figure 9 Optimization results for comparison analysis bracket………….......…...13 Figure 10 Brackets for comparison analysis…………………………….....………14 Figure 11 COMSOL results for designer bracket…………………………….…....14 Figure 12 COMSOL results for triangular truss bracket………..……………….....15 Figure 13 COMSOL results for Level Set bracket…………………………….…...16 v LIST OF SYMBOLS Open bounded set (--) Neumann boundary condition (--) Dirichlet boundary condition (--) A Isotropic Elasticity Tensor, (psi) Strain tensor, (in/in) u Displacement field, (in) f Vector valued volume forces, (lbf) g Surface loads, (lbf) Symmetric Matrix (--) Lame’s first parameter, (psi) Lame’s second parameter, (psi) tr Trace function, (--) I Identify matrix, (--) t Time, (sec) D Working domain, (--) V Fixed Volume, (in3) Lagrange multiplier, (--) Lagrange multiplier, (--) Optimization boundary, (--) H Mean curvature of boundary, (--) n Unit vector normal to surface, (--) Level set function, (--) vi ABSTRACT Sheet metal brackets are used extensively in the Aerospace industry to fixture components and plumbing to the structural shell of aircrafts or engines. Lightening holes are commonly utilized in sheet metal structures to minimize the weight of brackets. The shape and pattern of these lightening holes are usually left to the prerogative of the design engineer. Time constraints typically do not permit multiple iterations which can result in non-optimized designs. Shape and topology optimization by the level set method was reviewed as a tool to converge on the ideal lightening hole configuration. Through the use of the Scilab code developed by Karmann et al. [1], the level set methodology was tested on five sample brackets. The Level set method was able to successfully display its optimization capabilities and basic bracket construction norms could be derived from viewing the results. 1 1. Introduction 1.1 Background Since weight is a critical parameter in aerospace application, it is important to have lightweight but structurally sound brackets. Lightening holes are traditionally utilized to minimize weight of brackets. The shape and quantity of these holes is typically left up to the designer's discretion. This can lead to inefficient bracket designs. Traditionally, shape and topology of brackets is established by a manual iterative approach. A designer would make an initial guess, analyze the bracket, then adjust the shape accordingly. Recent developments utilizing the level-set method show promise to employ optimization codes to establish the shape and topology of brackets. Figure 1: Typical Sheet Metal Brackets [8] 1.2 Shape and Topology Optimization The goal of shape and topology optimization is to find the ideal structure where weight has been minimized and strength has been maximized. This is accomplished by iterating on the shape and topology of a structure until the model converges to the optimum arrangement. For structural shape and topology optimization, various methods have been studied. Four popular methods are the homogenization method, solid isotropic microstructure with penalty (SIMP) method, evolutionary structural optimization (ESO), and the level set method [2, 3]. 2 The homogenization method introduces a porous structure where the optimization is simplified by increasing or decreasing the size of the pores. This method is effective at deleting and adding holes in a structure but in more complicated structures, can result in something that is not manufacturable. Another downside to this method is that it can create intermediate densities [3]. The SIMP method was developed as an improvement to the homogenization method. This method penalizes regions of intermediate densities to converge on a more solid structure. The evolutionary structural optimization approach starts with an initial finite element model. Based on the specific objective function, the elements with the least contribution are removed. This method is very effective at obtaining a local optimized structure but may not achieve the global optimized structure. This project will review the use of the level set method in shape and topology optimization. The level set function was first established by Osher and Sethian to numerically track moving boundaries [4]. The function represents the structure as a moving boundary embedded in a higher order level set function. Complicated shape and topology changes are easily handled by moving the zero level set up or down within the higher order function as shown in Figure 2. Figure 2: Moving the zero level set [7] 3 1.3 Level Set Method using Scilab Code Scilab is a free, numerical computation software similar to Matlab. This project utilized the freely-distributed Scilab code developed by Karrman [1] based on the work from Allaire [4]. 4 2. Theory The approach used for this project was developed by Allaire [2] and independently verified by Wang among others [3]. The problem is set on an open bounded set, Ω, of real numbers. Dirichlet and Neumann boundary conditions are applied on the boundary ∂Ω. (1) (2) The displacement field u in Ω is the solution of (3) where A is defined from Hooke’s law and the strain tensor is defined by (5) (3) (4) (5) The objective function used in the Scilab code is the compliance with volume and perimeter constraints. The total amount of work done (the compliance) is (6) For a fixed volume, all admissible shapes are defined by Uad: (7) The optimization problem, with the addition of volume and perimeter Lagrange constraints, becomes the objective function minimization problem (10) The shape derivative is used to solve the optimization problem (11) To track the shape and topology of the structure, the level set function defined as 5 (12) is used. Time is introduced into the optimization code as an iteration parameter. At any time t, the boundary is represented by the zero level set. (13) By setting the volume equal to: V ( ' H Ae(u ) e(u )) the shape derivative (11) is used to solve (13). 6 (14) 3. Methodology 3.1 Level-Set Method This project used the level set method through the Karmann/Allaire Scilab code to search for the optimum bracket structure for various test cases. The optimum arrangement is dependent on the type of structure being analyzed, and the boundary loads and constraints. For this project, the goal is to minimize weight (Volume) and maximize strength. The code accomplishes this by sequentially creating linear elastic structures that when iterated on, minimize the objective function [2]. 3.2 Sample Brackets Five example brackets were selected with the goal of identifying and understanding key characteristics of an optimized bracket. Figure 3 shows these examples with their constraints and applied loads. The first three brackets were chosen because they are commonly seen examples in shape and topology optimization as well as readily available examples provided by Karmann/Alliare. The 4th and 5th brackets were chosen to get an appreciation for the effect of different loadings. The final bracket shown in Figure 3 is the bracket that was used for a comparison analysis. This bracket was chosen because it was not clear what the optimum structure should look like so a reasonable comparison of the Level Set capability could be achieved. 7 Figure 3: Bracket boundary conditions Example one is a simple cantilever beam with the left hand side fixed and a downward force applied to the center of the right hand side. Example two has a downward force applied to the center of the bottom edge. The bottom left hand corner is fixed in the x and y direction and the bottom right is fixed in the y direction. Example three has the left hand side fixed with opposing forces applied to the right hand side. A downward force is applied to the bottom right corner and an upward force is applied to the top right corner. Bracket example four has the left hand side fixed with two forces applied to the top right corner. One force is upward and the other is to the right. Example five is the same as example four except the horizontal force horizontal is moved to the lower right hand corner. The last bracket in Figure 3 has a downward force applied to the center of the bottom edge and a horizontal force applied to the top right corner. The bottom left hand corner is fixed in the x and y direction and the bottom right is fixed in the y direction. 8 4. Results and Discussion 4.1 Example Brackets Figure 4 shows the optimization process for example 1, the cantilever beam. The optimization starts with an initial guess of holes. One limitation of this code is the final structure is highly dependent on the initial guess. The code can easily handle combining holes and morphing the shape and topology but struggles to add new holes. If the initial guess isn’t ideal, it may not converge on the optimum solution [2]. Recent work on structural optimization using the level set function by Wang S. et al. shows this dependency can be minimized or eliminated by introducing radial based functions [7]. As can be seen through the results below, by the 6th iteration, the optimization code has almost converged on the final solution. This is also evident by viewing the convergence history that shows the objective function for each iteration. Figure 4: Optimization Results for Example Bracket 1 9 As can be seen in Figure 5, the optimal structure for example 2 looks like a bridge. Unlike example one, the convergence history shows the objective function actually increased at certain time steps. This is attributed to thin members breaking apart and combining to form larger holes. This shows the power of this optimization approach to overcome local minima and converge towards the global minimum. Figure 5: Optimization Results for Example Bracket 2 Example 3 shown in Figure 6 shows the level set optimization converged to a truss structure consisting of triangular members. As expected with the opposing loads at the top and bottom of the right side the final structure was symmetric. 10 Figure 6: Optimization Results for Example Bracket 3 Example 4 shown in Figure 7 shows the optimization code converged to a simple triangular shape. This example also shows the code was able to overcome a local minima and reduce the objective another 10% towards the final solution. Figure 7: Optimization Results for Example Bracket 4 Example 5 is shown in Figure 8. As with the previous examples, a truss structure with triangular elements comprised the optimized bracket. Example brackets 3, 4 and 5 were intentionally similar with the exception of the applied loads. This was done to gain an appreciation of how the structure changes with respect to the applied loads. In each example the quantity, thickness, and orientation of the structural members changed significantly to counteract the applied loads. 11 Figure 8: Optimization Results for Example Bracket 5 4.2 Establishment of Norms The five example brackets showed that a truss structure comprised mostly of triangular members is the optimum approach for designing brackets. This agrees well with established practice in structural engineering. In each example, the stiffening members generally aligned to form triangles to counteract the applied forces and distribute the stress evenly throughout the structure. One of the objectives of the project was to establish some basic norms for designing brackets, specifically lightening holes. The example brackets, specifically 3, 4 and 5, show a limitation of using established norms. Though it’s clear the optimum bracket should be comprised of a triangular truss structure, the arrangement of the stiffening members is highly dependent on the applied load. The norm of using triangular shaped lightening holes can not guarantee they will be applied correctly without using an optimization methodology like the Level set. 4.3 Bracket Comparison The comparison bracket, shown in Figure 3 was utilized to compare the capability of the code against a typical bracket design and another design based off the observation that a triangular truss structure is ideal. The goal was to confirm the capability of the level set methodology for shape and topology optimization. Figure 9 shows the results of the Scilab optimization. The number of iterations ran for this bracket was increased to 80 to ensure it converges on the optimum structure. The increase in the objective function 12 around the 50th iteration can be explained by the breaking of one of the structural members. The final structure was analyzed in COMSOL using a plane stress assumption. Figure 9: Optimization Results for Comparison Analysis Bracket 4.4 Stress Analysis using COMSOL COMSOL was used to compare the three brackets shown in Figure 10. These brackets represent three versions of the comparison bracket shown in Figure 3. The first represents a typical bracket design with a lightening hole to minimize weight. The second bracket represents a bracket created knowing the ideal design should be a triangular truss structure. The third bracket design was obtained using the level set method via the Karmann/Allaire Scilab code. The weight (volume) of the brackets were kept constant to independently review the structural capability. 13 Figure 10: Brackets for comparison analysis The total displacement results for the first bracket are shown Figure 11. As can be seen from the Figure, the largest displacement was concentrated near the load applied at the center of the bottom edge. Figure 11: COMSOL results for Designer Bracket The total displacement results for the bracket derived from the second bracket are shown in Figure 12. This bracket has additional stiffening members which helped counteract the load at the center of the bottom edge and distribute the displacement better throughout the bracket. 14 Figure 12: COMSOL results for triangular truss bracket Figure 13 shows the total displacement results for the bracket optimized using the Scilab code. As seen in this Figure, the displacement is well distributed throughout the bracket compared to the previous two examples. This is expected since the goal of the objective function is to minimize compliance. 15 Figure 13: COMSOL results for Level Set bracket Table 1: Summary of comparison study Displacement (in) Bracket 1: Designer Bracket 2: Triangular Bracket 3: Optimized Top, Right 0.00461 0.00114 0.00110 Bottom, Center 0.01486 0.00145 0.00118 Table 1 shows the summary of the displacement at the two load points. As expected, from a total displacement perspective, this first bracket did not fair as well as the others. The displacement at the bottom center was greater by a factor of 10. This underscores the importance of optimizing a bracket design for the applied loading conditions. The second bracket based of the triangular truss structure faired quite well when compared to the optimized bracket. The total displacement at top load was only 3% greater, however at the bottom load, the total displacement was 33% greater. This displays how a well structured optimization code can outperform any design standard by adjusting the location and thickness of the stiffening members to fit the specific problem. 16 5. Conclusions 5.1 Summary of Results The level set method was first tested on five example brackets. Through these examples, it was evident the level set method is capable of shape and topology optimization. From an initialization of round holes, the code quickly morphed the structure into a truss style bracket, arranging the stiffening members with respect to the applied loads and constraints. Through these example brackets, it was evident the ideal approach to minimize the weight of a bracket was to create a structure comprised of triangular truss members. This agrees with established structural engineering practices. A comparison analysis was performed on the three brackets shown in Figure 10 using COMSOL to test the capability of the Level Set method to minimize the objective function. As can be seen by comparing Figures 11, 12, and 13, the optimization code was able to evenly distribute the total displacement throughout the bracket structure. The code was also able to limit the total displacement at the load points better than the other two brackets. 17 References [1] Karrmann A. (2009). Structural optimization using sensitivity analysis Structural optimization using sensitivity analysis and a level-set method, in Scilab and Matlab [2] Allaire G., Jouve F., Toader A. (2004). Structural optimization using sensitivity analysis and a level set method, J. Comp. Phys., Vol 194/1, 363{393 [3] Wang M., Wang X. (2003). A level set method for structural topology optimization [4] Osher S., Sethian J. (1988). Front propogation with curvature-dependent speed: Algorithms based on Hamilton-Jacobi formulations, J Comp. Physics Vol 79 12-49 [5] Demers T. (2009). A Designer’s Approach for Optimizing an End-Loaded Cantilever Beam while Achieving Structural Requirements [6] Wang S., Lim K., Khoo B., Wang M. (2007). An extended level set method for shape and topology optimization, J Comp. Physics Vol 221 395-421 [7] Wang M. (2004). Physical Modeling and Optimization of Heterogenous Solid. Level Set Methods, 2004 International Conference on Manufacturing Automation. [8] Giddens. (2009). Giddens Capabilities. Retrieved from http://www.giddens.com/Capabilities.aspx 18