Optimization techniques play an important role in structural design, the very purpose of Brics"which is to find the best solutions from which a designer or a decision maker can derive a maximum benefit from the available resources." Optimization in Structural design By N. G. R. Iyengar It is fairly accepted fact that one of the most important human activities is decision making. It does not matter what field of activity one belongs to. Whether it is political, military, economic or technological, decisions have a far reaching influence on our lives. Optimization techniques play an important role in structural design, the very purpose of which is to find the best ways so that a designer or a decision maker can derive a maximum benefit from the available resources. The basic idea behind intuitive or indirect design in engineering is the memory of past experiences, subconscious motives, incomplete logical processes, random selections or sometimes mere superstition. This, in general, will not lead to the best design. The shortcomings of the indirect design can be overcome by adopting a direct or optimal design procedure. The feature of the optimal design is that it consists of only logical decisions. In making a logical decision, one sets out the constraints and then minimizes or maximizes the objective function (which could be either cost, weight or merit function). Structural optimum design methods can also be according to the design philosophy employed. Most civil engineering structures are even to-day designed on the basis of permissible stress criterion. However, some of the recent methods use a specified factor of safety against ultimate failure of the structure. Presently, the approach is based on the design constraints expressing the maximum probability of various types of events such as local or ultimate failure. The objective function is obtained by calculating each event and multiply it by the corresponding probability. The sum of all such products will be the total objective function. The constraints may also be probabilistic. These are suitable in situations when the loads acting on the structure are probabilistic or the material properties are random. During the early fifties there have been considerable advances in `art` and economy of the structural design through the use of better structural materials and refined knowledge of structural design processes. Thus, the aim was to put structural design on a scientific basis. The need for innovation and optimization arose in the challenging problems faced by the aerospace industry, which gave a Philip to research activities in this area. Requirements for Structural Design The basic requirements for an efficient structural design is that the response of the structure should be acceptable as per various specifications, i.e., it should at least be a feasible design. There can be large number of feasible designs, but it is desirable to choose the best from these several designs. The best design could be in terms of minimum cost, minimum weight or maximum performance or a combination of these. Many of the methods give rise to local minimum/maximum. Most of the methods, in general give rise to local minimum. This, however, depends on the mathematical nature of the objective function and the constraints. Optimization Problem The optimization problem is classified on the basis of nature of equations with respect to 41 maximized or minimized. For example, if this function is cost or weight, then the function is minimized. On the other hand if it is some other function, it is maximized. Structural Optimization Problem The structural optimization problem can be posed as: design variables. If the objective function and the constraints involving the design variable are linear then the optimization is termed as linear optimization problem. If even one of them is nonlinear it is classified as the non-linear optimization problem. In general the design variables are real but some times they could be integers for example, number of layers, orientation angle, etc. The behavior constraints could be equality constraints or inequality constraints depending on the nature of the problem. Minimum Weight Design of Structural Elements (Simultaneous failure mode theory) One of the earliest techniques employed for the optimization of structural elements is the Simultaneous Failure Mode Theory (SFMT). The approach has been employed to obtain Optimum Design (minimum strength to weight ratio) of elements like columns, plates, beams, cylinders, sheet-stiffener combination etc. The requirement for optimum design is that all the failure modes occur simultaneously for the possible design variables. As the number of design variables increase or the constraints on the behaviour variables increase, this approach does not lead to global optimum solution. A structural design problem can be represented as a mathematical model whose constituent elements are parameters, constraints and objective or merit function. The design parameters specify the geometry and topology of the structure and physical properties of its members. Some of these can be independent design parameters and others could be dependent on the independent design variables. Some of the design parameters are chosen by judgment and experience of the designer so as to reduce the size of the problem. This results in large savings in computational time, which inturn reduces the cost of the design. From the design parameters, a set of derived parameters are obtained which are defined as behaviour constraints e.g., stresses, deflections, natural frequencies and buckling loads etc., These bahaviour parameters are functionally related through laws of structural mechanics to the design variables. The objective or the merit function is formed by the proper choice of the design parameters. This function is either Minimize or Maximize F = F (x1,x2,x3……xn ) Subject to C1 = C1(x1,x2,x3……xn ) =0 C2 = C2 (x1,x2,x3……xn ) =0 . . Cn = Cn (x1,x2,x3……xn ) = 0 and f1 = f1 (x1,x2,x3,………xn) ³ 0 . . fn = fn (x1,x2,x3,……. .xn ) ³ 0 (1) (2) x1,x2,……xn are the design variables, C1,C2,….Cn are equality constraints and f1,f2,…..fn are the inequality constraints. The nature of the mathematical programming problem depends on the functional form of F,C and . If these are linear functions of design variables, then the mathematical programming problem is treated as linear programming problem. On the other hand if any one of them is a nonlinear function of the design variable, then it is classified as nonlinear programming problem. By and large most of the structural design problems belong to the later. A hyper surface in the design variable space, such that all designs represented by points on this surface are on the verge of failure in a particular failure mode for a particular load condition, it is called a behaviour constrained surface. Designs slightly to one side of the constrained surface will fail, while designs slightly to the other side will not fail in the particular mode and load condition associated with the specific behavior constrained surface. A hyper surface in the design variable space such that all designs represented by points on this surface are on the verge of being unacceptable for some external cause, not explicitly related to the behaviour constraint is called a side constraint surface. Methods of searching the best design can be 42 problem has both integer and real variables. In order to overcome this difficulty of mixed modes, optimization studies are carried out for multiple cell structure treating the number of cells as parameter rather than a design variable. On the basis of this study and other studies, one can categorically state that simultaneously failure mode theory does not lead to global optimum, when there is large number of behavior constraints. When two modes of failure simultaneously occur, one gets local minima. However, if there are only two modes of failure for a solution, minimization by parametric penalty function proves to be a better technique. (a) Optimization of Thin Walled Column Elements under Axial Load In aerospace structures and other sheet metal constructions, stiffeners, which are normally channel and Z-section, are used. The behaviour of these sections as individual elements differs from that of the sheet-stiffener combination. Without precisely understanding the behaviour of individual elements, we cannot understand the response of sheet-stringer combination. Investigations were carried out for-minimum weight design of columns of channel and Zsections. These two sections were chosen for the study since the behaviour of these sections under axial compressive load is very much different. For such members the number of behaviour constraints is large since the possible modes of failure are many. The objective function which is the weight of the column has three design variables. Sequential Unconstrained Minimization technique (SUMT) with interior penalty function approach is employed for optimization. As stated earlier this technique has an advantage that one is always in feasible domain and at every stage you get a better solution. The penalty function is minimized by using Fletcher-Powell method [5].The study once again brings out that, for the optimum design all the failure modes do not always occur simultaneously. In the case of Z-section overall buckling and local buckling of the flange are the critical modes at the optimum point. While for the channel section, the overall buckling and torsional buckling modes are active at the optimum point. Here again, the technique leads classified into simultaneous and sequential search. The simultaneous classification is characterized by the fact that all trial designs are selected before the analysis of any design is started. On the other hand the sequential search is characterized by the fact that future trial designs may be generated by using the results of the previous trial designs. Over the years a large number of techniques have been suggested to solve these equations resulting in an optimal design. However, these techniques do not always lead to a global optimum. These at best lead to local optimum. If the constraint equations and the objective function are convex functions, then it is possible to conclude that the local optimum will be a global optimum. However, in most of the structural design problem it is practically impossible to check the convexity of the function. One of the simplest ways is to start with different feasible solutions and check the solutions for global optimality. Optimization Activities at IIT Kanpur At this Institute, the work in the area of structural optimization started way back in 1968 in the departments of Aerospace, Civil and Mechanical engineering. Depending on the nature of the problem one picked up for the study, different methods of analyses have been employed like closed form solution, finite element and finite difference etc to obtain the behaviour constraints. The work was limited to isotropic materials and small size problems. With the development of computing facilities, large size problems were investigated. Since a structural designer is interested all the time in a feasible design, Eq. (1) is suitably modified through the introduction of the penalty term. This ensures that the design is always in the feasible domain. The way the penalty term was introduced depended on the problem. Some of the work during the period 1967-76 has been brought out in a book form by Iyengar and Gupta[1]. Katarya[2] considered the problem of optimization of multi-cellular wings designed from isotropic materials under strength and vibration constraints for simple loadings. The objective was to minimize the weight of the wing. Interior Penalty Function approach [3, 4] has been employed for optimization study. This 43 wing skin as an isotropic stiffened plate. Fig. 1 shows the wing structure idealization. Since at every stage of optimization the analysis has to be carried out, to save computational time, the minimum weight design is initially solved employing linearly approximated re-analysis. To assess the time saved by employing linearly approximated re-analysis, the same problem is solved using the exact analysis in the third and fourth unconstrained minimization. The study indicates that the time is reduced by 30 percent; however, the minimum weight is increased by 1.5 percent. The constrained optimization problem is solved as a sequence of unconstrained minimization problems by using the interior penalty function approach. Cubic interpolation method of one dimensional search, which makes use of the gradient, is used for finding the step length. This study reveals that it is possible to include multiple constraints for optimizing the structure. Initial parametric study carried out before optimization does result in reducing the number of design variables which in turn reduces the computational time. (c) Minimum Cost Design of Grid Floor The most common form of reinforced concrete construction of private and public buildings is Tbeam and grid floor. The design of these structures is generally based on two approaches; (I) stress design and (ii) strength design. It has been well established that the strength design is more logical and also economical. For the design slabs of various shapes and edge conditions limit design procedures have also been well established. These methods result in considerable economy in the design of reinforced concrete structures. However, one can further improve the design if one chooses the dimensions optimally. The cost of the structure is often a nonlinear function of the dimensions of the structure. It is necessary that the structure in addition to being low cost must meet the safety and functional requirements. These are also generally nonlinear. Adidam et.al [7] investigated the optimal design of T-beam and grid floors using Nonlinear Mathematical Programming Technique. The objective function here represents the cost of one beam and slab assembly per unit length along the beam span per unit spacing. This is also expressed as a ratio to a global optimum [1]. (b) Optimum Design of Wing Structure Fig.1: Wing Structure Idealization In real life problems, we encounter multiple constraints, both behaviour and side constraints. One such problem investigated and reported here is the optimum design of wing structure with design variables including aerodynamic parameters, such as sweep back angle, aspect ratio, and thickness to chord ratio, in addition to the usual geometric parameters [6]. Since number of variables is large, to minimize the computer time required for optimization, a parametric study of the behaviour quantities e.g., maximum deflection, stresses, buckling load and natural frequencies is needed to understand how these are influenced by the design variables. To start with, there are 13 design variables. This could be reduced to 9 without appreciably changing the behaviour constraints. Here again, the objective function is the weight of wing structure. The optimization problem is to reduce the weight satisfying the given constraints. As the geometry of the wing is quite complicated, it is not possible to obtain the behaviour constraints in the closed form. For the static and dynamic analysis, the wing is idealized by finite elements using constant stress triangular membrane elements and rectangular shear panels for the skin and web respectively. The stringers are represented by axial force members. Elastic buckling constraints are introduced by treating a typical portion of the 44 tank, as shown in Fig.2a. The base of the structure is subjected to ground acceleration during an earthquake. For the response analysis, the structure is idealized as a single-degree-offreedom system as shown in Fig. 2b. The stiffness of the system is computed by using the flexibility analysis. The objective function of the structure is the total volume of the structure. The lengths of the vertical members of the truss are treated as design variables with the condition that the sum of their lengths is a constant. Hence, instead of three design variables we shall have two. The total design variables are twelve. The constraints on the natural frequency of the structure is so specified that when the tank is partially filled the first few frequencies of the liquid oscillations are kept well below the natural frequency of the structure. This is done to avoid large amplitude liquid sloshing due to earthquake excitation. For the stress constraint, the ground acceleration during an earthquake is assumed to be stationary random process. For response calculation, the ground acceleration is locally considered white noise. The following observations can be drawn on the basis of this study: 1. Stress constraint is the only active constraint at the optimum point. The optimal design is sensitive to the way in which this constraint is defined. 2. The optimal design is sensitive to the degree of correlation between member stresses. (e) Minimum Weight Design of Sheet-Stringer Panels of cost per unit area of floor to the cost of one unit of concrete. An existing square grid of 18.83 meter span was optimized. This results in a relative cost of 58.76. Further, the optimal design turns out to be 1.2 meter square grid instead of existing one meter square. This indirectly results in saving of form work and material. Optimization Under Random Environment Most physical systems operate under random environment, e.g., flight vehicles subjected to gust loading, jet engine noise, boundary layer turbulence, trains, towers, buildings subjected to earth quakes. Nigam[8] and Narayanan[9] have applied the concepts of this design to structural optimization in random vibration environment. The problem is formulated by considering the time dependence of the response quantities and then reducing it to a standard nonlinear programming problem (NLP). The weight of the structure is optimized with constraints on natural frequencies, buckling stresses, geometric dimensions, and dynamic responses such as stresses, acceleration, and fatigue life of the structure. The constraints are expressed probabilistically. Choosing the probability of failure in such a failure mode is a matter of engineering judgment based on the functions of the structural system and on the possible consequences. (d) Elevated Water Tank Staging (Earthquake loads) 2(a) 2(b) Fig.3: Sheet-Stringer Combination Fig.2: Support Structure for Water Tank A typical sheet-stringer panel used in aircraft is Consider a truss structure supporting a water 45 shown in Fig.3. Instead of multiple panels we shall consider only a single panel between two stringers. In each frequency band, the lowest frequency corresponds to the stringer-torsion mode and the highest frequency to the stringer-bending mode. While designing for sheet-stringer panels, the distance between the frames is specified and the density of the material is constant. There are five design variables in this case. In the stringer-torsion mode, the adjacent panels vibrate out of phase, whereas in the stringerbending mode they vibrate in phase. The stringer-torsion mode does not get appreciably excited by the jet noise as compared with the stringer-bending mode. To reduce the response, the pressure spectrum should be so used that the frequency of the sheet corresponding to the stringer-bending mode is well above the frequency of the maximum sound energy. Fatigue damage is used as the other constraint. The problem is posed as an NLP and solved as an unconstrained minimization problem. The optimization study results in about 15 percent reduction in weight. Optimization Studies in Fibre reinforced Composites Fibre Reinforced Composite (FRP) materials are being employed as primary load carrying members in aerospace structures in view of the advantages they offer as compared to metallic structures. This is because, the fibre orientation in each lamina can be chosen depending on the designer's requirements. This results in changes material properties which are direction dependent unlike in metallic materials, which are direction independent. Substantial amount of composites are used in Light Combat Aircraft (LCA), Advance Light Helicopter (ALH) and the two seater trainer aircraft (HANSA). This has resulted in substantial savings in structural weight in-view of their high strength to weight ratio and high stiffness to weight ratio. The advantages can be further improved, provided these materials are used optimally. The author and his research students have contributed significantly to the literature on optimum design of FRP laminates. Most of these studies have been discussed in a book by Iyengar and Gupta[10]. These studies deal with the minimization of the weight of composite laminates subjected to various types of behaviour constraints. Optimization studies in composites is little more involved as compared to metallic materials as the number of variables increase substantially. Since the design variables will be a mix of real and integer variables this makes the analysis complicated. Genetic Algorithms for Optimization Studies The problem of mixed mode variables has been solved by the application of Genetic Algorithms (GA). These are suitable for complex optimization problems. Application of genetic algorithms for optimization studies is gaining wide interest because of their robustness in locating the global optimum. Recently this technique has been applied by Sivakumar [11] for the study of optimization of FRP laminates with and without cut-outs undergoing large amplitude oscillations. The cut-outs are of various sizes and shapes. In a recent paper, Sivakumar et.al., [12] have clearly brought out the advantages of GA for the optimum design of laminated composite plates with cutouts, over the conventional techniques, and also the effectiveness of GA in locating the optimum for problems involving large number of constraints and variables. They have concluded from the studies carried out, that the DFP method is not suitable for the problems considered. Finding an accurate result requires a large number of function evaluations than other techniques. The complex search technique finds the optimum solution with a small number of function evaluation than GA when the number of constraints is not large. When a large number of constraints are present, it takes a large number of function evaluation. The disadvantage is that it cannot handle discrete variables. GA seems to be the best tool to optimize composite laminates, since it can handle all types of variables providing the flexibility needed to solve such complex problems. Future Directions in optimization Studies So far the investigations have been confined to optimization with single objective function. Design of a complex system like space vehicles, aircrafts etc., requires a large number of merit functions to be satisfied for the best design. 46 Furthermore, optimum design of sub-systems does not lead to optimal design of the entire system. For example, in the case of aircrafts, the objective function will be to minimize the weight. However there could be other objectives like minimizing the Drag, maximizing the range etc., Problem of this type has to be treated as multi objective function optimization. The work in this direction has already been initiated. REFERENCES [1]Iyengar, N.G.R and Gupta S.K ` Programming methods in structural design`,Edward Arnold Pub.Ltd.U.K.1980 [2]Katarya, R.` Optimization of multi-cellur wings under strength and vibrational constraints for simple loading, M.Tech thesis,IIT Kanpur ,1973 [3]Fiacco,A.V and McCormick,G.P`The sequential unconstrained minimization technique for nonlinear Programming: Aprimal-Dual method ,Management sc. 10,360-366,1964 [4]Fiacco,A.V.andMcCormick,G.P `SUMTwithout parameters, Operations research,15,820-827,1967 [5]Fletcher,R and Powell, M.J.D., ` A rapidly convergent descent method for minimization, Computer J. 6,163168,1963 [6] Rao,V.R. Iyengar, N.G.R. and Rao,S. S,Optimization of wing structures to satisfy strength and frequency requirement,` Computers and Structures,10,669-674,1977. [7]Adidam, S.R., Iyengar,N.G.R. and Narayanan,G.V.` Optimum design of T-beam and grid floors,J. structural engineering,6,113-124,1978 [8]Nigam,N.C. Structural optimization in random vibration environment,AIAA.J.10,551-553,1972 [9]Narayanan,S., Structural optimization in random vibration environment, Ph.D Thesis, IIT Kanpur, 1975. [10]Iyengar,N.G.R.and Gupta,S.K.Structural design Optimization,Affiliated East west press Ltd. New Delhi, 1997 [11]Sivakumar,K., Optimum design of laminated composite plates with dynamic constraints,Ph.D thesis IIT Kanpur [12] Sivakumar,K. and Iyengar, N.G.R and Deb Kalyanmoy, Optimization of composite laminates with cutouts using genetic algorithm, variable metric and complex search methods, Eng. Opt.32,635 –657,2000 About the author: Dr. N.G.R. Iyengar is a Professor in the Department of Aerospace Engineering. He did his Ph.D. at IIT Kanpur and has served the institute in many capacities for the past 37 years. He has played the lead role in establishing the ARDB center of excellence for composite structures and technology in our institute. His research interests include structural analysis and optimization, composite structures, composite materials and their failure. 47 48