# One challenge in the application of control systems to

## Comments

## Transcription

One challenge in the application of control systems to

INTEGRATED DEVICE PLACEMENT AND CONTROL DESIGN IN CIVIL STRUCTURES USING GENETIC ALGORITHMS Ping Tan1, Shirley J. Dyke,2 Andy Richardson3, and Makola Abdullah4 ABSTRACT One challenge in the application of control systems to civil engineering structures is appropriate integration of a control system into a structure to achieve effective performance. Placement of control devices is strongly linked to the performance of a control system, and the most appropriate device placement scheme is strongly dependent on the performance objectives of the control system. Additionally, for the most effective control system, the placement scheme should be integrated with the design of the controller rather than sequential. This paper proposes an integrated technique to place devices and design controllers based on the use of genetic algorithms. The approach is flexible, allowing the designer to base the placement scheme on performance goals and/or system requirements. Active control devices are used and an H2/LQG controller based on acceleration feedback is selected for this study based on previous successes with this approach in civil engineering systems. To illustrate the proposed methodology, two numerical examples are considered. The first example considers a 40-story shear building, and the second is a full-scale, irregular, 9-story building. Control is achieved through the placement of one or more active control devices placed on various floors in an active bracing configuration. The improvements in the effectiveness of the proposed methodology as compared to previously developed techniques are demonstrated through comparative studies. Keywords: active control, genetic algorithms, optimal placement, protective systems 1. Postdoctoral Research Assoc., Dept. of Civil Engineering, City College of the City University of New York (research performed while a Postdoctoral Researcher at Washington University in St. Louis) email: [email protected] 2. Edward C. Dicke Professor of Engineering, Dept. of Civil Engr., Washington University in St. Louis, St. Louis, Missouri, 63130, email: [email protected] (Corresponding Author). 3. Structural Engineer, PBS&J, 5300 West Cypress Street, Suite 200, Tampa, Florida 33607, email: [email protected] 4. Associate Professor, Dept. of Civil Engr., Florida A&M University, Tallahassee, FL 32210, email: [email protected] 1 Tan et al. INTRODUCTION To improve the safety and serviceability of structures excited by dynamic loads, significant effort has been directed toward the possibility of employing protective systems in civil engineering structures (Housner et al., 1997; Spencer and Sain, 1997; Spencer and Nagarajaiah, 2003). To sustain the design and improve the performance of these systems, the development of methods for optimal utilization of these types of devices is an important research topic. One of the challenges of this technology is in achieving the most effective performance of these systems while considering cost through appropriate integration of the control system into a structure. The capabilities of the control system strongly depend on the placement of the control devices. Researchers have been investigating techniques for device placement since the introduction of the concept of structural control. Chang and Soong (1980) placed a limited number of active devices in a structure for modal control by minimizing a performance index. Lindberg and Longman (1984) discussed the appropriate number and placement of devices based on independent modal space control. Vander Velde and Carnignan (1984) considered structural failure modes to place the devices, and Ibidapo (1985) used structural mode shapes as optimality criteria for device placement. Cheng and Pantiledes (1988) computed the controllability index associated with each story of a building, from which the actuator locations were provided. Most of these methods do not explicitely relate placement to performance and the methods are often problem specific. Search algorithms have also been utilized for device placements. Takewaki (1997) used a gradient-based approach to search for an optimal solution that would minimize a desired system transfer function. Teng and Liu (1992) and Xu and Teng (2002) developed an incremental algorithm for placement of active/passive control devices. Wu et al. (1997) investigated both iterative and sequential approaches to placement of passive devices in irregular structures. Zhang et al. (1992) and Lopez et al. (2002) proposed sequential search methods for determining the optimal 2 Tan et al. damper placement. Although these algorithms are computationally efficient, they may results in local optima being found (Goldberg, 1999). As an alternative, a guided random search method, the simulated annealing method, was employed to place devices (Chen et al. 1991; Liu et al. 1997). Although this method has an advantage over heuristic-based algorithms in seeking global optima, it does not always provide efficient solution techniques. Another versatile global optimization approach is to use genetic algorithms (GA). This approach has shown significant promise in its ability to solve problems where the objective function is not a continuous function of the design variables and the variable space is discrete (Goldberg, 1999). Singh and Moreschi (2002) examined placement of passive energy dissipation devices in a multistory building for seismic protection. Wongprasert and Symans (2004) employed GA for identifying the optimal damper distribution to control the nonlinear seismic response of a 20-story benchmark building. Simpson and Hansen (1996) used GAs to optimize the placement of actuators to actively control interior harmonic sound levels in a cylinder with floor structure. Ponslet and Eldred (1996) employed a GA approach to design an isolation system. Abdullah et al. (2001) combined genetic algorithms and a gradient-based optimization technique to design the optimal position of direct velocity feedback control controllers in buildings. Li et al. (2001) have developed a multi-level genetic algorithm to solve a multi-tasking and multi-model optimization problem simultaneously. In practice, a placement design procedure should be simple and efficient, and require only an acceptable amount of computational effort. Control devices should be placed in the structure with due consideration to the designer’s design objectives, including performance goals such as response reduction, as well as system resources (number of devices, force capacity, actuator dynamics, device stroke/velocity, etc) and cost. Additionally, the placement strategy should allow other information such as actuator dynamics and disturbance characteristics to be integrated into the design procedure. To appropriately account for all of these issues, the placement strategy should be 3 Tan et al. integrated with the controller design. Thus, a combinatorial optimization problem can be considered in which control algorithm parameters and the device locations are determined simultaneously. However, in nearly all of the previous placement studies identified, controller design and device placement are performed separately. Generally this may not result in the most effective and efficient control design, although in certain situations the optimal control system, or a control system with similar performance, may be identified. This paper demonstrates the use of genetic algorithms (GA) for integrated placement of control devices and controller design. This integrated approach is shown to improve upon some previously developed techniques to place devices. It is also demonstrated to be flexible, allowing the designer to base the placement scheme on a variety of performance goals and system resources. Additionally, the designer could easily incorporate actuator dynamics and characteristics of the disturbance, if known, into the design procedure. For this study, the structure is assumed to remain linear, and active control devices are employed. As shown by Yoshida and Dyke (2004), this assumption is often appropriate even in the case of a structure exhibiting nonlinear behavior. An H2/LQG controller based on absolute acceleration feedback is selected. To validate the proposed approach two numerical examples are presented in which the proposed approach is employed to place active control devices and the system performance is evaluated. In both examples, control is achieved through the placement of devices in an active bracing configuration. The numerical results also indicate the methodology is simple, effective and efficient. METHODOLOGY Integrated Design and Device Placement Procedure Genetic algorithms is a heuristic random search technique based on the concept of natural selection and natural genetics of a population. Thus, GA is a population-based method for searching large combinatorial design spaces to identify a globally optimal or near-optimal combination of 4 Tan et al. design variables (Goldberg, 1999). Although there is no guarantee that the optimal solution will always be obtained, the robust behavior and random nature of methods based on GAs make them attractive for this application. Genetic algorithms typically require the use of coded strings which represent an underlying set of parameters. These parameters are coded into a chromosome, which has an associated fitness value determined by the selected performance function of interest. The corresponding fitness function is used to reflect the degree of ‘goodness’ of each chromosome. In each generation, the fitness of each chromosome is determined, and a new generation is developed by combining features of chromosomes with the highest fitness values. GA proceed by reproducing chromosomes to search for those with increasing fitness. In searching for the best design, GA work with a population, which is comprised of different chromosomes simulating natural evolution by means of random genetic changes that produce successively better approximations, ultimately leading to a final design solution. The three basic operations included in GA are selection, cross-over and mutation. Selection determines which individuals are chosen for mating and how many offspring each selected individual produces. Crossover and mutation operations create new designs for further exploration in the search space. The rules for mating, crossover and mutation are defined in a probabilistic manner. As a new population is selected, the performance function is evaluated for each new design to determine its fitness. Reproduction, crossover, and mutation are subsequently applied to successive populations. The power of GA lies in their robustness as well as the efficiency and simplicity of the technique. GA are especially suitable for problems in which the fitness function is not a continuous and differentiable function of the design variables or the variable design space is discrete. GA are employed here for solving a mixed optimization problem, in which both the positions of control devices and controller gains are to be optimized. The proposed integrated approach 5 Tan et al. uses GA to simultaneously place the devices and design the controller (i.e., select the weighting matrices). Figure 1 provides a flow chart of the proposed design procedure. Note that the third column, describing the steps involved in evaluating each string, is the primary contribution of the proposed method as compared to previously proposed GA techniques for civil structures. To begin, an initial population of feasible designs (Old-Gen) is randomly generated. In the context of placement of control devices, a feasible design is a specific distribution of devices, and a chromosome represents a possible arrangement of control devices. The initial population is selected according to their fitness for production of offspring, using both cross-over and mutation operations. These offspring are inserted into the population, replacing the parents and producing a new generation. In evaluating each chromosome of a population, a search is conducted using a specified objective function to determine the control gains. Each control design is evaluated by calculating its performance criterion, and the selection operations are subsequently implemented. The strings which yields the best fitness, designated ‘Best-Gen’ are kept, and the procedure is repeated. With each successive generation, the ‘Best-Gen’ population is selectively updated. The selection, breeding and updating process continues until the specified number of generation, i is exceeded. The string that yields the best fitness value is then identified. A limitation of GA is that there is no guarantee that the optimal solution is always obtained. However, in practice, and particularly in the problem considered herein, efficiency of an algorithm yielding a near-optimal solution often supersedes the need for the optimal solution. For example, if a designer can obtain a placement yielding near-optimal performance (within a few percent of the fitness of the optimal solution) in two hours of CPU time, it would not be worthwhile to spend two weeks of searching all possible combinations to find the optimal solution. Thus, GA provide an attractive tool for control system design. 6 Tan et al. Control Design Consider an actively controlled structure excited by a seismic input. Assuming that the controller provides adequate control action to keep the structure itself in the linear range, the equation of motion can be written as Mx·· + Cx· + Kx = – MΓ x·· g + Λu (1) where M , C and K are the mass, damping and stiffness matrices of the system, respectively, x is the vector of relative displacements, Γ describes the influence of the ground excitation, x·· g is the ground acceleration vector, u is the vector of control forces, and its coefficient matrix, Λ , is the matrix determined due to the location of control devices. Defining the state vector T T T x s = [ x x· ] , this equation can be written in state space form as x· s = Ax s + Bu + Ex·· g (2) z z z z = C x s + D u + F x·· g (3) y y y y = C x s + D u + F x·· g (4) where A and B are the system matrices, z is the regulated output vector, which is obtained from z z z the mapping matrices, C , D and F . Similarly, y is the measurement vector, which is obtained y y y from the mapping matrices, C , D and F . For control design, the input excitation x·· g may be modeled as a filtered, stationary white noise. The disturbance shaping filter can be written in state space form, and the states of the system model can be augmented accordingly (Dyke et al., 1996a,b). If the number of states in the structural model is high, controllers can also be designed using a reduced-order system model. An H2/LQG controller is selected to reduce the responses of the structure based on its performance in previous applications (Yi et al., 2001; Dyke et al., 1995, 1996a–c, 1998). In this ap- 7 Tan et al. proach, an infinite horizon performance index is selected to weight the vector of regulated outputs given above by z τ T 1 J = lim --- E ∫ ( z Q 0 τ→∞ τ T z + u R u ) dτ (5) where R is an identity matrix, and Q = qI is a weighting matrix with equal weighting placed on each of the regulated outputs. Thus we can design controllers by adjusting the weighting factor q and the selected regulated outputs. Further, the ground excitaion and measurement noise are assumed to be independent, and the measurement noises are identically distributed. For the designs herein, the ratio of the power spectral densities is taken to be γ = S wj wj ⁄ S vi vi = 25 , where w j is the jth term of the disturbance vector and v i is the ith term in the measurement noise vector. The control and estimation problems can be performed separately, according to the separation principle (Stengel 1986; Skelton 1988), yielding a controller of the form u = – Kx̂ s (6) where x̂ s is the Kalman filter estimate of the state vector, and K is the full state feedback gain matrix. The Kalman Filter optimal estimator is given by y y x̂ s = Ax̂ s + Bu + L ( y – C x̂ s – D u ) (7) where L is the observer gain matrix of the stationary Kalman filter. Calculations to determine K and L were performed using the MATLAB routine lqry.m and lqew.m within the control toolbox. Determination of Control Gain To integrate the controller design with device placement, for each string the controller is designed by adjusting the coefficient of the regulated response weighting matrix, q, while considering a specific objective. The objective function on which the design is based can be stated in terms of the control resources (i.e., forces) needed or the system performance (i.e., response reduction). 8 Tan et al. These two approaches for determining the control gain are designated as approach 1 and 2, respectively. The corresponding objective functions used to determine the optimal control gain are defined by J g1 = max ( σ fk ) ≤ [ f l ] for k = 1, 2, …, r (8) where σ f k is the RMS force of the kth device and [ f l ] is the RMS device force limit, and maxσ z ij i, j J g2 = -------------------- ≤ [ zl ] maxσ uzij for i = 1, 2, …, n , j = 1, 2…, m (9) i, j where σ zij is the controlled value of the specified regulated RMS response of the jth frame at ith floor, and σ uz ij is the corresponding uncontrolled response. Here [ z l ] is the chosen level of response reduction. The response used in Eq. (9) may be absolute acceleration, relative displacement or interstory drift, depending on the control objective. RMS responses are determined by solving the Lyapunov equation (Soong and Grigoriu, 1993) and the excitation is assumed to be a stationary filtered white noise. For purposes of illustration, assume that the second objective function is used. The objective function declines monotonously with increasing weighting coefficient until a specified level, or threshold, in performance is achieved. Once this is achieved, an objective function value (i.e., sum of all RMS forces) is assigned to that string, and we continue on to the next string. Each string requires a different control force to achieve a set performance level. The string using the smallest control force, represents the best control device placement scheme. Alternatively, if the first objective function had been selected, the threshold would be determined by evaluating the maximum control force required for each design, and the selected objective function value would correspond to the selected response quantity. Both approaches are adopted and demonstrated in the numerical examples presented herein. If the threshold value is set too aggressively, it may not 9 Tan et al. be possible to reach that threshold, typically because the controller is unstable. Thus experience, or prior knowledge of the system, is helpful. Performance Function The fitness of a particular arrangement of control devices has an associated performance function. Performance functions are used by the GA to evaluate the reduction in certain response quantities of design interest or control resources used, and make appropriate comparisons of the resulting designs. Note that the performance function selected should not correspond to the same responses used in the objective function for control gain determination. Using the same response for both would lead to circular logic in design of the controller. In this paper, the following performance criteria are considered. The first criterion is a displacement based criterion, and is given as maxσ z ij i, j J 1 = -------------------maxσ uz ij i, j for i = 1, 2, …, n , j = 1, 2…, m (10) in which σ zij is the controlled RMS relative displacement of the jth frame at ith floor, and σ uz ij is the corresponding uncontrolled relative displacement. The second criterion is an acceleration based criterion, given as maxσ ·· z ij i, j J 2 = -------------------maxσ ·· for i = 1, 2, …, n , j = 1, 2…, m (11) uz ij where σ ·· is the controlled RMS accelerations of the jth frame at ith floor, and σ z ij ·· uz ij is the corre- sponding uncontrolled absolute acceleration. The third criterion based on the interstory drifts is given as maxσ d ij i, j J 3 = -------------------maxσ ud i, j ij for i = 1, 2, …, n , j = 1, 2…, m 10 (12) Tan et al. where σ dij is the controlled RMS interstory drift of the jth frame at ith floor, and σ ud ij is the corresponding uncontrolled interstory drift. An additional criterion based on the control force is J 4 = sum ( σ u k ) for k = 1, 2, …, r (13) where σ uk is the kth RMS control force. A majority of the time in optimization techniques based on GAs is typically spent on the evaluation of the fitness function for each individual. The fitness of a particular arrangement of control devices is defined as a function of the associated performance function values. RMS responses and control forces are determined using the Lyapunov equation (Soong and Grigoriu, 1993). The GA codes using in this analysis are implemented in MATLAB (Pohlheim, 1999). NUMERICAL EXAMPLES In the following sections, two numerical examples considering full-scale buildings are presented to demonstrate the proposed approach and to assess the achievable performance in each. When available, comparisons of the proposed technique to that obtained by other researchers are presented. Although placement of sensors is generally an important step in the design of a controller, it is not a focus of this research effort. Thus, in each design considered herein, the absolute acceleration of several floors of the example buildings are used for feedback. Various control objectives are considered for the designs discussed herein. Some controllers are designed to minimize the relative displacements and interstory drifts of each floor by equally weighting the corresponding regulated responses of each floor. In the design of other controllers, the control objective is to minimize the absolute accelerations of each floor by equally weighting the absolute accelerations. Values for these performance functions, J 1 – J 4 , are also obtained to evaluate the resulting designs for each example. Control designs which are open-loop unstable are eliminated due to the fact that they cannot be implemented. Additionally, although it would be relatively simple to incorporate actuator dynamics into the placement scheme as discussed in Fig. 1, 11 Tan et al. and would yield improved results in real-world applications (Dyke et al., 1995), we neglect actuator dynamics to focus on validation of the proposed approach. Example 1: 40-story Shear Building This example employs a 40-story, shear-building model, selected from Abdullah et al. (2001). The structural properties of the each floor of the structure are: mass m i = 1290 ton; stiffness 6 k i = 10 kN/m; and damping constant c i = 100 kN/(m/s). Controllers are designed using a reduced-order model considering only the first eight modes, but are evaluated using the full system model. A Kanai-Tajimi disturbance shaping filter (Soong and Grigoriu, 1993) is used for the design and the parameters were selected as ω g = 17 rad/sec and ς g = 0.3 (Ramallo et al., 2002). In this example, 10 actuators are to be placed. Each actuator has a maximum RMS control force of 4000 kN. In situations in which this limit is achieved, several smaller devices could be placed in parallel to achieve this force. Each generation contains 25 strings, and 200 generations are considered. Table 1 provides a comparison of the results of three search algorithms designed to minimize the relative displacement, and provides the associated system performance and resources required. Cases A and B correspond to the results obtained using two approaches described in Abdullah et al. (2001). Case C corresponds to the results of the proposed methodology. The variation of the performance and resulting positions of devices with generation of evolution are shown in Fig.2. The results indicate that, for a given level of RMS control force, significant improvements in the control performance can be achieved when devices are placed using an integrated procedure. Note that the total control forces required for each design are quite close. In Cases A and B some devices are not using their full capacity, while a small number of control devices achieve most of the control action. However, a more uniform distribution of RMS control forces can be 12 Tan et al. achieved with the proposed procedure. Thus, the proposed methodology allows for the capacities of devices to be fully utilized in real-world applications, resulting in a more cost-effective design. Example 2: 9 Story Irregular Building For a regular, symmetric building as discussed in Example 1, the control performance is primarily dependent on the floors where actuator are placed, and the placement within the floors is not critical to the performance. However, in the case of a building with an irregular plan (in which the lateral and torsional motions are coupled), the number of potential control device locations is considerably larger, making the problem more complex. Not only the floor, but also the placement within the floor, are essential variables to consider for achieving effective performance in an irregular building. The challenges and opportunities regarding the control of irregular structures are discussed in more detail by Yoshida (2003), Yoshida et al., (2003) and Yoshida and Dyke (2005). Example 2 considers the full scale, 9 story irregular building used by Yoshida (2003) and Yoshida and Dyke (2005) for semiactive control studies (Obayashi, 2002). The main structural system of this building is steel reinforced concrete (SRC) and the plan view is shown in Fig. 3. On each floor there are five bays in the x-direction and four bays in the y-direction. An important feature of this building is that the distribution of shear walls makes this structure irregular, demonstrating coupling the lateral and torsional motions. A two-dimensional, linear, lumped-parameter model of this building is employed to capture and focus on the x-translation and the torsional responses of the building (Yoshida, 2003; Yoshida and Dyke, 2005). The first two calculated natural frequencies are 0.83 Hz (translation in x-direction), and 1.29 Hz (torsion). The damping is assumed to be 2% for all modes. The parameters required to generate a lumped parameter model of this structure (including mass, moment of inertia, lateral and torsional stiffnesses) are provided in Table 2 for each floor along with the resulting frequencies of the model. Control devices can be installed within each frame on each floor, for a 13 Tan et al. total of 45 viable device locations. For feedback to the controller, the absolute acceleration of each floor is measured at Frame A and Frame E for a total of 18 acceleration responses. To provide a basis for comparison, the performance of the control designs resulting from implementation of the proposed approach are compared to four possible placement schemes of 9 devices. The designs and results are shown in Table 3 for both drift- and acceleration-based designs. Because it is not reasonable to compare the results using GA to all combinations of device placements, Cases D1–D4 and A1–A4 are several placement schemes selected for comparative purposes based on practical experience, in which cases the devices are placed along the edges of the lower storys. Cases D5 and A5 correspond to the proposed scheme with the device RMS force limit set to 150 kN. Note that placement schemes D5 and A5 perform significantly better than the other cases in Table 3. These results indicate that the proposed approach yields a superior control system compared to the other arrangements for this irregular building. Thus, appropriate placement of the devices in the building is important for the design of a high performance, cost-effective solution, especially in the case of an irregular structure. Implementation Studies As discussed previously, the proposed integrated approach can be implemented for a variety of design goals. For instance, the designer may have a specific performance goal to achieve, or a specific force/cost of the system. Next a few sample design scenarios are considered to demonstrate the variety of approaches that can be employed. The irregular structure studied in Example 2 is used for these implementation studies. First we consider the results of the proposed technique when implemented using specific numbers of control devices. Table 4 shows the resulting placements for various device numbers for both drift-based and acceleration-based designs. Observe that as the number of devices increases, the results do not always include the locations in the previous case with fewer devices. For example, when one device is placed using acceleration weighting, the optimal location is po- 14 Tan et al. sition 26, but the most effective placement scheme using three devices does not include position 26. Although there is no guarantee that we have achieved the global minimum in the evaluation criteria, this implies that a sequential design approach does not generally yield the most effective placement because the addition of each device changes the system dynamics. Next we examine the effect of varying the force limit. The results for placement of 9 devices using various force limits are provided in Table 5. The placement of devices were determined for several chosen levels of RMS control forces. At a low level of force, devices tend to be placed at the same position. However, when the RMS force limit is increased, the devices are spread throughout the structure. Clearly a larger control force than the chosen force level is required at the position in which more than one actuator is clustered, and a single device cannot provide the required force if the limit is too low. Moreover, most of the actuators are concentrated at frame A (i.e., locations 1, 6, 11, 16, 21, 26, 31, 36, or 41) when the force limit is low. With an increase in allowable control force, the proposed approach placed devices at other frames to more effectively reduce both torsional and translational motions. Another case is investigated to identify the placement scheme to achieve specific, quantitative performance objective. Table 6 shows the performance and selection of nine device locations for various levels of response reduction. Type A controllers focus on minimizing interstory drifts and type B controllers are based on minimizing absolute accelerations. This study provides an example to study the relationship between system cost (i.e., number of devices required) and performance. To demonstrate the performance of the proposed methodology for real earthquakes, the El Centro and Northridge earthquakes are used. The Northridge record is used to determine the number of devices because this earthquake requires larger control forces. Using the total control force required at each control device location to realize the performance, the number of control devices 15 Tan et al. which have a capacity of 1,000kN are determined, as shown in Table 7. Note that the number of devices roughly doubles for each increase in performance level. Structural responses for the two earthquakes are provided in Table 8 for all six controllers along with the corresponding percent of the uncontrolled value in parenthesis. The results indicate that the type A controllers can reduce drifts more effectively than the type B controllers, while the type B controllers can reduce accelerations more efficiently than the type A controllers. Controller 3A and 3B achieve the best performance at the expense of higher control forces. However, note that all controllers can reduce both drift and acceleration responses substantially for the two earthquakes. CONCLUSIONS An integrated technique for control device placement and controller design has been proposed and demonstrated. The approach employs genetic algorithms to identify an optimal or nearoptimal solution to the placement problem. In addition, the approach was shown to be flexible, allowing the designer to base the placement scheme on performance goals and/or control system resources (i.e., system costs). This offers the benefit of yielding a more practical solution in which the standard sized (i.e., off-the-shelf) control devices can be implemented with known force capacities and there is a higher degree of uniformity in the forces required by the various devices. Active control devices in conjunction with H2/LQG controllers were considered and acceleration feedback was used in all control designs. The effectiveness of the proposed methodology has been demonstrated through numerical studies and, when possible, the results were compared to those of other researchers and placement schemes based on practical experience. The results of these numerical studies also show that the most effective distribution of actuators is highly dependent on the desired level of performance and the device force limit. Furthermore, the methodology allows for the designer to consider the trade-off between economy and performance. 16 Tan et al. ACKNOWLEDGEMENT This research is partially supported by National Science Foundation Grant CMS 97–33272 (Grant Monitor, Dr. Shih-Chi Liu). Additionally the authors would like to thank Dr. Osamu Yoshida of Obayashi Corporation for the 9-story full scale building model used herein. REFERENCES Abdullah, M., Richardson, A. and Hanif, J., (2001), “Placement of Sensors/Actuators on Civil Structures Using Genetic Algorithms.” Earthquake Engineering and Structural Dynamics, Wiley, Vol. 30, pp. 1167–1184. Chang, M.I.J. and Soong, T.T. (1980). “Mode Control of Multistory Structures.” Journal of Engineering Mechanics Division, ASCE, Vol. 102, pp. 613–23. Chen, G.S., Bruno, R.J. and Salama, M. (1991). “Optimal Placement of Active/Passive Control of Flexible Structures”. AIAA Journal, AIAA, Vol. 29, No. 8, pp. 1327–1334. Cheng, F.Y. and Pantelides, C.P. (1988). Optimal Placement of Actuators for Structural Control. Technical Report NCEER-88-0037, Multidisciplinary Center for Earthquake Engineering Research, Buffalo, New York. Dyke, S.J. Spencer Jr., B.F., Quast, P., and Sain, M.K. (1995). “Role of Control-Structure Interaction in Protective System Design,” Journal of Engineering Mechanics, ASCE, Vol. 121, No. 2, pp. 322–338. Dyke, S.J., Spencer Jr., B.F., Quast, P., Sain, M.K., Kaspari Jr., D.C. and Soong, T.T. (1996a). “Acceleration Feedback Control of MDOF Structures,” Journal of Engineering Mechanics, ASCE, Vol. 122, No. 9, pp. 907–918. Dyke, S.J., Spencer Jr., B.F., Quast, P., Kaspari Jr., D.C., and Sain, M.K. (1996b). “Implementation of an AMD Using Acceleration Feedback Control,” Microcomputers in Civil Engrg., Special Issue on Active and Hybrid Structural Control, Blackwell Publishers, Cambridge, MA, Vol. 11, pp. 305–323. 17 Tan et al. Dyke, S.J., Spencer Jr., B.F., Sain, M.K. and Carlson, J.D. (1996c). “Modeling and Control of Magnetorheological Dampers for Seismic Response Reduction,” Smart Materials and Structures, Institute of Physics, Vol. 5, pp. 565–575. Dyke, S.J., Spencer Jr., B.F., Sain, M.K., and Carlson, J.D. (1998). “An Experimental Study of MR Dampers for Seismic Protection,” Smart Materials and Structures: Special Issue on Large Civil Structures, Institute of Physics, Vol. 7, pp. 693–703. Goldberg, D. E. (1999). Genetic Algorithm in Search, Optimization and Machine Learning. Addison Wesley: Reading, MA. Housner, G. W., Bergman, L.A., Caughey, T. K., Chassiakos, A. G., Claus, R. O., Masri, S. F., Skelton, R. E., Soong, T. T., Spencer, Jr. B. F., and Yao, J. T. P., (1997). “Structural Control: Past, Present and Future.” Journal of Engineering Mechanics, ASCE, Vol. 123, No. 9, pp. 897–971. Ibidapo, O. (1985). “Optimal Actuators Placement for the Active Control of Flexible Structures.” Journal of Mathematical Analysis and Applications, Elsevier, Vol. 105, No. 1, pp. 12–25. Li, Q. S., Liu, D. K. and Zhang, N., Tam, C. M. and Yang, L. F. (2001). “Multi-level Design Model and Genetic Algorithm for Structural Control System Optimization,” Earthquake Engineering and Structural Dynamics, Wiley,Vol. 30, No. 6, pp. 927–942. Lindberg, R. E. Jr., and Longman, R. W. (1984). “On the number and Placement of Actuators for Independent Modal Space Control.” Journal of Guidance, Control and Dynamics, AIAA, Vol. 7, No.2, pp. 215–221. Liu, X. J., Begg, D. W. and Matravers, D.R. (1997). “Optimal Topology/Actuator Placement Design of Structures Using SA.” Journal of Aerospace Engineering, ASCE, Vol. 10, No. 3, pp. 119–125. 18 Tan et al. Lopez, G. and Soong, T.T. (2002). Efficiency of a Simple Approach to Damper Allocation in MDOF Structures. Journal of Structural Control, Wiley, Vol. 9, No. 1, pp. 19–30. Obayashi Corporation Technical Research Institute, Internal Report, 2002; 63024–6. (in Japanese, not publically available). Pohlheim, H., (1999). “Evolutionary Algorithms: Overview, Methods and Operators.” Documentation for: Genetic Evolutionary Algorithm Toolbox for Use with Matlab version: toolbox 1.92 documentation 1.92, Mathworks, Inc. Ponslet, E. R. and Eldred, M. S. (1996). “Discrete Optimization of Isolator Locations for Vibration Isolation System.” Proc. of the AIAA, NASA, and ISSMO Symposium on Multi-Disciplinary Analysis and Optimization, Technical Papers, Part II, pp. 1703-16. Ramallo, J. C., Johnson, E. A., and Spencer, B. F.(2002) " ‘Smart’ Base Isolation Systems", Journal of Engineering Mechanics, ASCE, Vol.128, No. 10, pp. 1088-1099. Singh, M.P. and Moreschi, L.M. (2002). “Optimal Placement of Dampers for Passive Response Control,” Earthquake Engineering and Structural Dynamics, Wiley, Vol. 31, pp. 955–976. Simpson, M.T. and Hansen, C.H. (1996). “Use of Genetic Algorithms to Optimize Vibration Actuator Placement or Active Control of Harmonic Interior Noise in a Cylinder with Floor Structure.” Journal of Noise Control Engineering, Vol. 44, No.4, pp. 169–184. Skelton, R.E. (1988). Dynamic Systems Control: Linear Systems Analysis and Synthesis. John Wiley & Sons, New York. Soong, T.T. and Grigoriu, M. (1993). Random Vibration of Mechanical and Structural Systems, Prentice Hall, New Jersey. Spencer Jr., B.F. and Nagarajaiah, S. (2003). “State of the Art of Structural Control,” Journal of Structural Engineering, ASCE, Vol. 239, No. 7, pp. 845-56. 19 Tan et al. Spencer Jr., B.F. and Sain, M.K. (1997). “Controlling Buildings: A New Frontier in Feedback,” IEEE Control Systems Magazine: Special Issue on Emerging Technologies (Tariq Samad Guest Ed.), Vol. 17, No. 6, pp. 19–35. Stengel, R.F. (1986). Stochastic Optimal Control: Theory and Application. John Wiley & Sons, New York. Takewaki, I. (1997). “Optimal Damper Placement for Minimum Transfer Function.” Earthquake Engineering and Structural Dynamics, Wiley, Vol. 26, No. 11, pp. 1113-1124. Teng, J. and Liu, J. (1992). “Control Devices Optimal Installment for the New Vibration Control System of Multi-structure Connection.” Proceedings of 10WCEE, Madrid, Spain, 19-24 July, 1992, Balkema, Rotterdam, pp. 2069-072. Vander Velde, W. E. and Carignan, C. R. (1984). “Number and Placement of Control System Components Considering Possible Failures.” Journal of Guidance, Control and Dynamics, AIAA, Vol. 7, No. 6, pp. 703-709. Wongprasert, N. and Symans, M. D. (2004). "Application of a Genetic Algorithm for Optimal Damper Distribution within the Nonlinear Seismic Benchmark Building," J. of Engineering Mechanics, ASCE, Special Issue on Structural Control Benchmark Problems for Dynamically Excited Buildings, Vol. 130, No. 4, pp. 401-406. Wu, B., Ou, J.-P., and Soong, T.T. (1997) “Optimal Placement of Energy Dissipation Devices for Three Dimensional Structures,” Engineering Structures, Elsevier, Vol. 19, pp. 113–125. Xu, Y. L. and Teng, J. (2002). “Optimal Design of Active/Passive Control Devices for Tall Buildings under Earthquake Excitation.” The Structural Design of Tall Buildings, Wiley & Sons, Vol. 11, No. 2, pp. 109-127. Yi, F., Dyke, S.J., Caicedo, J.M., and Carlson, J.D. (2001), “Experimental Verification of MultiInput Seismic Control Strategies for Smart Dampers,” Journal of Engineering Mechanics, ASCE, Vol. 127, No. 11, pp. 1152-1164. 20 Tan et al. Yoshida, O. (2003). Torsionally Coupled Response Control of Earthquake Excited Asymmetric Buildings: Development and Application of Effective Control Systems Using Smart Dampers, Doctoral Dissertation, Washington University in St. Louis, St. Louis, Missouri, USA. Yoshida, O., Dyke, S.J., Giacosa, L.M. and Truman, K.Z. (2003). “Experimental Verification on Torsional Response Control of Asymmetric Buildings Using MR Dampers.” Earthquake Engineering and Structural Dynamics, Vol. 32, No.13, pp. 2085–2105. Yoshida, O. and Dyke, S.J. (2004). “Seismic Control of a Nonlinear Benchmark Building Using Smart Dampers,” Journal of Engineering Mechanics: Special Issue on Benchmark Control Problems, ASCE, Vol. 130, No. 4, pp. 386-392. Yoshida, O. and Dyke, S.J. (2005) “Response Control in Full Scale Irregular Buildings Using MR Dampers,” Journal of Structural Engineering, ASCE (in press). Zhang R. H. and Soong T. T. (1992). “Seismic Design of Viscoelastic Dampers for Structural Application.” Journal of Structural Engineering, ASCE, Vol. 118, No. 5, pp. 1375-1392. 21 Tan et al. Start Start Form Model (with Actuator Dynamics or Disturbance Char.) Place Sensors Reduce Model Apply GA Evaluate Final Design Old-Gen Old-Gen Crossover Crossover Mutate Mutate New-Gen New-Gen Evaluate Evaluate Each Each String String Selection Selection Find Find Q Q LQG Design Compute perforance criterion Mutate Mutate Best-Gen Best-Gen Update Update P<i End End Figure 1. Flow Chart Describing the Design Procedure. 22 Tan et al. Story J1 Generation Generation (a) (b) Figure 2. Example Evolutionary Curves for the Proposed Methodology (a)Performance and (b) Position of Devices. 23 Tan et al. D 17,050 X CM 13,200 C B A 1 2 3 5 4 25,600 E 6,000 6,600 6,000 7,000 34,400 6,400 6,400 6,400 7,600 Y 7,600 6 Figure 3. 9-Story Irregular Building (Plan View). 24 Tan et al. Table 1: Performance Comparison for Placement Schemes. J4 (1,000kN) q 0.4763 0.5992 28.65 109.05 123489 12 14 26 36 0.4213 0.5621 28.42 109.2 11111 22223 0.3535 0.5638 28.63 109.4 Case Device Locations A 3 4 7 10 11 15 19 24 30 36 B C J1 25 J2 Tan et al. Table 2: Structural Parameters (x-Direction). Total Moment of Stiffness Inertia Floor Weight i Wi Ii (kN) (kgcm ) 9 15556 2.50E+12 9206 2.65E+10 8 10198 1.64E+12 11516 3.07E+10 k xi 2 Torsional Stiffness k θi (kN/cm) (kNcm/rad) 7 10118 1.63E+12 11408 3.25E+10 6 10205 1.64E+12 11731 3.32E+10 5 10295 1.66E+12 12257 3.48E+10 4 10294 1.66E+12 13182 3.83E+10 3 10382 1.67E+12 15172 4.35E+10 2 10470 1.68E+12 16301 4.63E+10 1 10983 1.77E+12 18309 5.34E+10 26 Tan et al. Table 3: Comparison of Placement Schemes. Interstory Drift Based Designs Case Device Locations J3 q D1 112233678 0.3962 1012.6 D2 1 1 5 5 6 6 10 10 11 0. 3968 1012.3 D3 1 1 6 6 11 11 16 16 21 0.2917 1012.8 D4 26 35 35 36 36 41 41 42 42 0.5601 1011.9 1 6 6 11 16 20 22 26 33 0.2463 1012.8 D5 * Acceleration Based Designs Case Device Locations J2 q A1 112233678 0.4598 107.7 A2 1 1 5 5 6 6 10 10 11 0.3969 107.8 A3 1 1 6 6 11 11 16 16 21 0.3966 108.2 A4 26 35 35 36 36 41 41 42 42 0.4822 107.4 A5* 5 10 14 16 20 21 22 26 31 0.2523 108.2 *. Designed using proposed approach. 27 Tan et al. Table 4: Placement and Performance for Various Numbers of Devices. Interstory Drift Based Designs No. of Devices Device Locations J3 q 1 16 0.7628 1011.5 3 11 16 21 0.4839 1012 6 1 6 11 16 21 26 0.3224 1012.5 9 1 6 6 11 16 20 22 26 33 0.2484 1012.8 1 2 6 8 11 14 17 19 22 25 26 31 0.1936 1013.1 12 Acceleration Based Designs No. of Devices Device Locations J2 q 1 26 0.7665 106.9 3 16 21 31 0.4923 107.4 6 10 11 16 25 26 26 0.3322 107.8 9 5 10 14 16 20 21 22 26 31 0.2523 108.2 1 5 5 8 8 15 17 20 21 21 28 31 0.2284 108.4 12 28 Tan et al. Table 5: Placement and Performance for Various Force Limits. Interstory Drift Based Designs J g1 (100kN) Device Locations J3 q 1 6 6 11 11 17 17 21 21 27 0.315 1012.4 1.5 1 6 6 11 16 20 22 26 33 0.246 1012.8 3 1 4 6 10 12 17 23 26 31 0.150 1013.7 5 1 5 8 12 18 23 29 31 38 0.119 1014.15 Acceleration Based Designs J g1 Device Locations J2 q 1 10 11 11 20 21 21 26 30 31 0.332 107.6 1.5 5 10 14 16 20 21 22 26 31 0.252 108.2 3 3 4 5 9 12 17 20 23 28 0.189 109 5 3 4 4 5 9 11 17 24 29 0.154 109.55 (100kN) 29 Tan et al. Table 6: Placement and Performance for Various Response Reductions. Interstory Drift Based Designs J g2 Controller Device Locations J4 (kN) q 70 1A 15 16 16 16 16 17 21 21 21 202 1010.8 50 2A 11 16 21 21 26 26 26 26 26 427 1011.5 30 3A 1 6 6 6 11 16 21 21 26 881 1012.5 (%) Acceleration Based Designs J g2 Controller Device Locations J2 q 70 1B 11 11 26 26 31 31 36 41 41 200 106.2 50 2B 6 16 16 21 21 26 26 36 36 430 106.9 30 3B 3 10 11 1121 21 26 26 30 101 107.9 (%) 30 Tan et al. Table 7: Position of Control Devices. Position Controller 1A Number 15 16 17 21 – – 1 6 2 6 – – Position Controller 2A Number 11 16 21 26 – – 4 4 9 20 – – Position 1 6 11 16 21 26 Controller 3A Number 10 25 12 11 17 10 Position Controller 1B Number 11 26 31 36 41 – 4 4 5 3 4 – Position 6 16 21 26 36 – Controller 2B Number 5 8 8 10 10 – Position 3 10 11 21 26 30 Controller 3B Number 10 11 20 20 23 15 31 Tan et al. Table 8: Responses Due to Historical Earthquakes. Configuration Acceleration (m/s2) Interstory Drift (m) Maximum RMS Maximum RMS Total Control Force (1,000kN) Maximum RMS El Centro Earthquake, N-S Component, Imperial Valley on May 18, 1940 Uncontrolled 0.0350 0.0112 10.42 2.58 – – Controller 3A 0.0154(0.44) 0.0028(0.25) 8.203(0.79) 1.278(0.50) 30.6 5.52 Controller 2A 0.0247(0.71) 0.0051(0.45) 7.567(0.73) 1.300(0.50) 12.5 2.76 Controller 1A 0.0307(0.88) 0.0074(0.76) 9.533(0.92) 1.854(0.72) 4.93 1.33 Controller 3B 0.0208(0.60) 0.0028(0.37) 5.532(0.53) 0.797(0.31) 24.0 6.18 Controller 2B 0.0242(0.70) 0.0057(0.50) 7.034(0.68) 1.278(0.50) 11.6 2.69 Controller 1B 0.0317(0.91) 0.0086(0.77) 8.154(0.78) 1.887(0.73) 5.25 1.22 Uncontrolled 0.0140 0.0946 28.1379 3.6263 – – Controller 3A 0.0555(0.59) 0.0056(0.40) 17.879(0.64) 1.781(0.49) 74.1 9.33 Controller 2A 0.0825(0.87) 0.0089(0.63) 22.030(0.78) 2.202(0.61) 34.5 4.48 Controller 1A 0.0905(0.96) 0.0103(0.80) 27.369(0.97) 2.826(0.85) 12.53 1.77 Controller 3B 0.0834(0.88) 0.0086(0.61) 13.893(0.49) 1.390(0.38) 69.2 10.84 Controller 2B 0.0795(0.84) 0.0097(0.70) 19.010(0.68) 2.084(0.57) 31.0 4.33 Controller 1B 0.0867(0.92) 0.0120(0.86) 23.518(0.84) 2.698(0.74) 14.71 1.86 Northridge Earthquake, N-S Component, Sylmar on January 17, 1994 32 Tan et al.