Semi-Active Control of Structures Equipped with FDSAB Using Integrated Genetic Algorithm-Fuzzy Controller and Neutral Networks

Transcription

1 Internatonal Journal of Emergng Engneerng Research and Technology Volume 3, Issue 3, March 25, PP 3-24 ISSN (Prnt) & ISSN (Onlne) Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated Genetc Algorthm-Fuzzy Controller and Neutral Networks Hosen Ghaffarzadeh Department of Cvl Engneerng Faculty of Structural Engneerng, Unversty of Tabrz, 29 Bahman Blvd., Tabrz, Iran ABSTRACT In ths paper, an ntegrated fuzzy rule-based control strategy for buldng structures ncorporated wth sem actve frcton system wth amplfyng braces (FDSAB) subected to earthquake exctaton was developed. A genetc algorthm (GA) s used for optmzaton of the fuzzy logc controller. The man purpose of employng a GA s to determne approprate fuzzy control rules as well to adust parameters of the membershp functons. To mprove the performance of control strategy n reducng tme delay between the response measurement and the control acton, an expertly traned artfcal neural network was constructed based on the nformaton of sesmc structural response aganst an earthquake exctaton whch obtaned from several dynamc tme hstory analyss of a structure. The effcency of the developed system s demonstrated by the numercal smulaton of a seven storey shear frame confgured wth FDSAB devces at each storey. The controlled response of the frame was compared wth results obtaned from the classcal lnear quadratc regulator (LQR) method. The results showed the proposed control strategy was hghly effectve n reducton of both the sesmc response and tme delay effect n control scheme. Keywords: Sem actve control system, Fuzzy controller, Genetc Algorthm (GA), Artfcal Neural Network (ANN), Regulator Lnear Quadratc (LQR), Tme delay, Frcton System wth Amplfyng Braces (FDSAB) INTRODUCTION Structural control strateges have been a subect of ntensve research and have been recognzed as a promsng approach for response suppresson n recent decades. Generally, several vbraton control technologes have been developed. Structural control systems can be classfed to actve, sem-actve and passve control methods. Sem-actve structural control has receved consderable attenton snce they possess some remarkable advantages over passve and actve systems. In fact, they not only mantan the relablty of passve control systems, but also provde the versatlty of actve control systems wth a much lower power requrement whch s of a great mportance n structural control durng earthquakes []. The control effectveness of structural systems s hghly dependent on the control strategy used for desgnng sem-actve control law. Conventonal control algorthms are relant on havng an accurate model of the system ncludng clpped-optmal algorthm [2] optmal controllers [3, 4, 5] Lyapunov stablty theory [6] skyhook controllers and contnuous sldng mode (CSM) controllers [7] decentralzed bang-bang, maxmum energy dsspaton [8] lnear quadratc regulator (LQR) [9] and others of that lk. Although these model-based strateges have been successful suppresson of structural vbratons, they suffer from some nherent shortcomngs for structural applcatons and ther performance s strongly affected by the accuracy of the model selected. The other approaches, as alternatve to the classcal control algorthms, conssts of methods whch do not rely on a system model. They are based on the actual measured responses of the system. Ths category ncludes neural network and fuzzy control methods [,, 2]. Because of nherent robustness, easness to handle the uncertantes, nonlneartes and heurstc knowledge, fuzzy logc control (FLC) theory has attracted the attenton of researchers for vbraton control of structural systems. Fuzzy control systems have been successfully appled to wde varety of *Address for correspondence ghaffar@tabrzu.ac.r Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25 3

2 Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated control problems. A sem-actve fuzzy control strategy for sesmc response reducton usng a magneto-rheologcal (MR) damper has been proposed by Cho et al [3]. Fuzzy-logc control strategy for protectng buldngs aganst severe exctatons usng magneto-rheologcal (MR) dampers has been wdely nvestgated [4, 5, and 6]. Genetc algorthms have been appled as effectve search technques to many felds of optmzaton problems. In structural control applcatons, GA can be utlzed as an optmzaton technque to pursue the ultmate goal of reducng the structural responses that determnes the structural safety. In desgn of FLC, choosng approprate membershp functons s a sgnfcant and tme-consumng part. Some valuable efforts were acheved towards the applcaton of a genetc algorthm (GA) to the desgn of a FLC [6, 7, 8, and 9]. In ths manner, the GA optmzed FLC s used to adust the parameters of the fuzzy membershp functons and fndng approprate fuzzy control rules. The reason s ther smplcty, ease of operaton, mnmal requrements and parallel and global perspectve. In recent years, neural network technques have been wdely used for the vbraton control n cvl engneerng [2, 2]. Neural networks are powerful computatonal tools may also be traned to predct the modal response of the structures, and then obtan the requred control force to be appled to the structure. Neural-network-based control algorthms have the promse of evolvng nto powerful adaptve controllers. A hybrd fuzzy logc and neural network controller was proposed to mprove the control performance of multple-nput multple-output control systems [22]. Applcaton of a neuroncontrol method for sem-actve vbraton control of stay cables usng MR dampers [] showed that the proposed control strateges could effectvely mplement sem-actve vbraton control of stay cables wth the use of MR dampers. Although prevous researches have made full use of the advantages of the neural network and fuzzy logc controller, few researches have adopted both tools to resolve the tme-delay and nstantaneous choosng of control parameters n sem-actve control or actve control. A neural network as a controller, act as estmator n whch, control force needed for the next tme step s fully determned from the nformaton avalable at the current tme. Consequently, the tme delay assocated wth the control algorthm s elmnated. Ths beneft of ANN controller can be wdely mplemented n the control strategy of structures. ZhaoDong and Yng Qng [23] developed an ntegrated control strategy n whch the tme-delay problem was solved by a neural network and the control currents of the MR dampers were determned quckly by a fuzzy controller. Among many sem-actve control devces, sem-actve frcton systems wth amplfyng braces (FDSAB) are more attractve to use because of mechancal smplcty, ther small sze and low operatng power requrement. Also, the control forces requred when FDSAB s used are smaller compared wth structures controlled by frcton dampers connected ether to chevron or dagonal braces [24]. Ths makes FDSAB more effectve to use n structural control especally for structures that may be subected to severe earthquakes. In ths paper, a combned applcaton of genetc algorthm (GA), fuzzy logc controller (FLC) and artfcal neural network (ANN) s adopted for sem-actve control of buldng frames usng FDSABs. In order to desgn an accurate fuzzy logc controller, membershp functons and fuzzy rules were optmzed usng the GA optmzer. An expertly traned ANN model s used to reduce tme delay between the response measurement and the control acton. The ANN was constructed based on the nformaton of sesmc structural response aganst an earthquake exctaton whch obtaned from several dynamc tme hstory analyss of a structure. In a numercal example, the developed FLC s appled to a ten storey buldng and tme hstory analyses are conducted to evaluate the performance. To verfy effectveness of the developed strategy, a lnear quadratc Regulator (LQR) controller s desgned and consdered n the smulatons for comparson purposes. The desgn process developed n ths paper can be used to desgn new structures or for upgradng of exstng buldngs. FRICTION DAMPING SYSTEM WITH AMPLIFYING BRACES Rbakov and Gluck [25] proposed an actve FD connected between the chevron braces and the top floor daphragms. The results ndcated FDs can effectvely suppress the vbratons of a structure n both cases. Sem-actve Frcton Dampers (FDs) were descrbed and tested expermentally by Nshtan et al [26]. Over the past few years, frcton dampers have been wdely nvestgated as an energy dsspaton system to mtgate the structural response due to dynamc loadng [, 27]. Several varable frcton dampers have been conceptually nvestgated and some have been mplemented. As noted earler, FDSAB system s more effectve to use as control system than smlar sem-actve 4 Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25

3 Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated frcton dampers connected ether chevron or dagonal braces due to lower amount of control forces requred. Consequently, n ths study, FDSAB system s used to vbraton control of structure. The other feature of FDSAB devce s ts large amount of dsspated energy at each cycle of loadng. The frcton dampng system wth amplfyng braces (FDSAB) s a sem actve devce conssts of a square frame, wth bars connected by hnges, a system of cables and a sem-actve frcton damper as shown n Fgure. The proposed frcton damper conssts of an external cylnder, two nternal half-cylnders and a pneumatc camera.the damper s connected to the top and bottom floor daphragms by means of cable. In FDSABs, the ar pressure n the pneumatc camera of each damper adusts the slp force level accordng to the optmal soluton yeldng the requred frcton forces based on control algorthm. Typcal force dsplacement hysteretc loop for a frcton damper has rectangular shape. Assumng that the cable length remans constant, the amplfed dsplacement transferred to the FD at each floor level takes the form: d FD, Where: E * 2 * b 4 * b c () d b * F 2 2 FD, FD, 2b, c * 2 B st, 2 d cos d cos 2. The energy dsspated at the th floor s gven by: Where F FD, s the control force at the th floor. In current study, a seven story frame equpped wth FDSAB system s desgned to nvestgate the effcency of the proposed control strategy n such systems. LINEAR QUADRATIC REGULATOR CONTROL METHOD The classcal lnear quadratc regulator algorthm (LQR) has been extensvely used for sem actve control. Ths control method requres that all the values of the state varables to be avalable. In ths algorthm, the control force u(t) s determned by mnmzng the performance ndex over duraton of exctaton. A dscrete-tme lnear system s defned by: x(k + ) = Ax(k) + Bu(k) (3) In ths system, the optmal control force takes the form u(k) = Gx(k). G s a m 2(n + m) feedback gan matrx where n s number of stores and m s the number of FDSAB systems appled to control the buldng responses. Ths optmal control force s obtaned by mnmzng the performance ndex gven by [9]: (2) J = k (x T (k)qx(k) + u T (k)ru(k) ) (4) Where the symmetrc weghtng matrces Q and R are the weghtng matrces to be appled to the response and control energy respectvely. Q s a 2(n + m) 2(n + m) postve sem-defnte matrx; and R s a m m postve defnte matrx. The control force u(k) s weghted n the performance ndex to allow regulaton wthout usng excessve control energy. Control force vector u(k) regulated by state vector x(k) s determned as: u (k) = - G x(k) = -((R T - T + B PB) B PA)x(k) (5) And P = Q + A T (P - PB(R + B T PB) - B T P)A (6) Where G and P are the soluton of the classcal dscrete tme algebrac Rccat equaton. The choce of sutable Q and R for a specfc problem s usually dffcult and requres experence and engneerng nsght. Here, weghtng matrces, P and Q are consdered as: -2 R = I Q = I (7) Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25 5

4 Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated CONTROL STRATEGY Fgure. Constructon scheme of FDSAB Devce The proposed control strategy ntegrates fuzzy logc and Genetc Algorthm and employs artfcal neural network as an effectve analyzer as llustrated n Fgure 2. The controller conssts of two parts. One s fuzzy logc controller (FLC), whch calculates the control forces; the other s the artfcal neural network (ANN) part whch s used as the predctor for controller to reduce tme delay of computng control forces. Interacton relatonshp between the structural responses and control output sgnals s dffcult to determne by conventonal methods. In ths manner, the FLC can be effectvely used for the control problem to establsh a correlaton between a set of nputs and outputs. Dsplacement and velocty response of system are used as the nputs of a fuzzy logc controller. The fuzzy controller output sgnals are propagated to control devce n order to generate dampng forces that mtgate the system response. Approprate desgn of a FLC s an mportant part of a control scheme. In an ordnary control method, the membershp functons and assocated rules of a fuzzy controller are usually determned by tral and error. For effcency, an optmal desgn of fuzzy control rules and membershp functons of the fuzzy controller s desred. In ths study GA s employed as an effcent approach to optmally desgn a fuzzy logc controller. Fuzzy controller parameters such as the dstrbuton of the membershp functons and fuzzy rules are updated from ntal membershp functons and fuzzy rules by GA's powerful searchng and self-learnng ablty. The ntegrated GA-FLC uses GA to produce an effectve fuzzy rule base that establshes a proper correlaton between selected nput and output of the FLC. In order to solve nherent tme-delay problems that exst n tradtonal control strateges, the ntegrated GA-FLC s moblzed by artfcal neural network to acheve a powerful control scheme, n whch ANN acts as predctor to mnmze response tme of the control devce. Tme delay s elmnated by computng the state of structure n the proposed GA-ANN-FLC. Fgure2. Integraton of fuzzy logc, genetc algorthms and neural network for control strategy usng FDSAB 6 Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25

5 Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated NUMERICAL STUDY A seven story shear frame confgured wth FDSAB devces per floor was selected to nvestgate the performance of the ntegrated sem-actve control strategy n reducng structural responses. The confguraton of uncontrolled and controlled frame s llustrated n Fgure 3. The structural matrces are: M kg K Nm.2.2 In order to evaluate the adaptablty of the proposed strategy, the model of the structure s subected to sesmc motons. Exctaton records that are used for numercal smulaton are El Centro SE (94) and Kobe NF7 (995). Fgure 4 llustrates ground acceleraton tme hstory for the selected earthquakes. The results of the example frame responses controlled by the combned scheme are compared wth those controlled by LQR control methods. Fgure3. Seven-story structure (a) uncontrolled, (b) wth FDSAB Fgure4. Ground motons used n ths study Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25 7

6 Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated Desgn and Optmzaton of the FLC FLC s employed to descrbe a complex mappng between a set of nputs and a set of outputs. The desgn of the sem-actve fuzzy controller nvolves specfcaton of the response quanttes to be used as nput to the fuzzy controller and the dstrbuton and type of membershp functons to be used for the selected nput varables and the defnton of the output varables. The FLC structure takes velocty and dsplacement of stores as an nput and provdes control force as an output. Input varables are defned by seven membershp functons defned on the normalzed unverse of dscourse [, ]. Output varable havng eleven membershp functons are also defned on the normalzed unverse of dscourse [-, ]. A reasonable range of nput values must be selected for the nput membershp functons snce f the range s too large or too small, the outermost membershp functons wll rarely or essentally be used, respectvely, and thus lmt the varablty of the control system. The membershp functons chosen for the nput and output varables are common trangular shaped n genetc fuzzy systems, as llustrated n Fgure 6. The defntons of the fuzzy varables of nput membershp functon are as follows: NL = Negatve Large, = Negatve Medum, = Negatve Small, =Zero, = Postve Small, = Postve Medum and PL = Postve Large. The defntons of the fuzzy varables of the output membershp functon are as follows: NVL= Negatve Very Large, NL = Negatve Large, = Negatve Medum, = Negatve Small, NVS = Negatve Very Small, ZE =Zero, PVS = Postve Very Small, = Postve Small, = Postve Medum and PL = Postve Large, PVL = Postve Very Large. Snce normalzed unverses of dscourse were used for both nputs and for the output to the fuzzy controller, scalng factors were requred to map the varables to these domans. In ths paper 2, 2 and 4 was selected as constant scalng factors of dsplacement, velocty and control force, respectvely, whch were obtaned accordng to control of the frame usng LQR method. The fuzzy rule base s determned to represent the relatonshp between nput and output fuzzy varables, where the output vares n proportonal to the scale of each gven nput. The rule base module s constructed by specfyng a set of f -premse-then-consequent statements. For example, the multple-nput multple-output IF THEN rules of the fuzzy control are shown n the form: R Then : If x s A y s B and and and and x y p m s A p s B m (8) where R denotes the -th rule of the fuzzy nference rule, =,2,...,q, x, x2,..., xp are the nputs of the fuzzy controller, A s the lngustc value wth respect to x of rule, y, y2,..., ym are the B outputs of the fuzzy controller and s a fuzzy sngleton functon defned by experts. The fuzzy rule-base for all possble f -premse-then-consequent statements can be lsted n a tabular form, whch s referred to as a fuzzy rule table. The defuzzfcaton module, the last component of the fuzzy logc, operates on the fuzzfed outputs obtaned from the nference mechansm. As noted earler, we adopt the center of gravty (COG) defuzzfcaton method n ths study. Fgure5. Input and output membershp functons of fuzzy controller In FLC desgn, defnng several parameters such as membershp functons s a knowledge-based process whch s obvously not an optmal soluton. GA optmally establshes a reasonable fuzzy correlaton between the selected structural responses and the correspondng control forces provded by FDSAB. Consequently, GA s employed as an adaptve method for optmzng the FLC system accordng to a ftness functon that specfes the desgn crtera n a quanttatve manner. The 8 Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25

7 Dsplacement Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated maxmum dsplacement response of the example frame due to earthquake exctaton s taken as obectve of the optmzaton problem, whch should be mnmzed. Therefore, ths maxmum dsplacement response s used n the ftness functon. In ths stage, to desgn the GA-FLC, the fuzzy membershp functons and rule base n fuzzy system are adusted. In order to adust the membershp functons used for nput and output varables, 3 parameter of P()-P(3), as shown n Fgure 5 are consdered, n whch usng the GA optmzer, the performance of the desgned FLC system can be optmzed. Due to symmetry, parameters of only half of the nput and output membershp functons, descrbed earler, have been consdered as desgn varables. Fgure 6 shows the fnal optmzed membershp functons for nput and output vectors. The parameters of nput and output membershp functons should be optmzed whle the rule base remans unchanged. After optmzaton of membershp functons, the rules of the fuzzy controller are optmzed based on the fnal membershp functons. Optmzed rule base table of fuzzy controller s tabulated n Table. Proper GA operator parameters are very mportant n mprovng the GA tournament. These parameters such as populaton sze, crossover rate and mutaton rate are chosen accordng to the problem's type. These parameters are gven n Table 2 for membershp functons and rule base optmzaton. The constrant of convergence s consdered as generatons of the populaton. Table. Optmzed rule base of fuzzy controller NL PL Fgure6. Optmzed Input and output membershp functons NL PVL PL PVS PL NVS NVS Sesmc Response Mtgaton of GA-FLC Controller Velocty PVS NVS PL R NVS NL NVL To verfy the control desgn valdty and the effectveness of proposed method, the results of the example frame responses controlled by the genetc algorthms and fuzzy logc controller (GA-FLC) method are compared wth those controlled by LQR algorthm whle the uncontrolled system response s used as the base lne. In ths stage, tme delay effect s not consdered and neural network estmator s not ncluded n the controller. Fgure 7 provdes the results of controlled dsplacement response of 7th floor of the example frame calculated by the GA-FLC compared wth the correspondng uncontrolled ones. It can be seen that GA-FLC results n sgnfcant reducton of roof dsplacement due to sesmc motons. The maxmum response reducton s about 62% and 56% for Elcentro and Kobe earthquakes, respectvely. The effcency of the proposed control strategy s further evaluated by comparson wth results of LQR control method n Fgure 8. As can be observed, GA-FLC s more effectve than the tradtonal LQR algorthm n dsplacement response mtgaton. GA-FLC results n about % and % maxmum response reducton more than LQR for Elcentro and Kobe earthquakes, respectvely. Furthermore, Fgure 9 provdes control force of 7th floor needed for obtanng the response reductons n the GA-FLC system. It can be seen that the requred control force n GA-FLC control scheme s more than that of the LQR system. However, more dsplacement suppresson n the GA-FLC system leads to more reducton n member sze of the desgned buldng Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25 9 PVS PVS NL

8 DISP (cm) DISP (cm) DISP (cm) DISP (cm) Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated whch s of a great mportance from the economcal pont of vew. Another noteworthy pont that should be taken nto consderaton s that, as noted earler, LQR control method requres that all the values of the state varables to be avalable. Ths means that the electronc sensors must be nstalled n all stores of the buldng whch mpose extra expenses. 5 STORY7-DISPLACEMENT-ELCENTRO 4 STORY7-DISPLACEMENT-KOBE UNCONTROLLED GA - FLC 3 UNCONTROLLED GA-FLC Fgure7. Comparson of controlled roof dsplacement of the example frame calculated by the GA-FLC system wth the uncontrolled ones Desgn of the Neural Network In order to reduce tme delay between measured response and control acton n the proposed scheme, the control forces must be calculated based on applyng state, not measured state. Consequently, the GA-FLC controller s mproved wth ANN to modfy the nput state vector that has been placed before the fuzzy controller. In other words, the man goal of the neural network s to predct the dynamc responses of the structure as nputs to the fuzzy controller to determne control force of FDSAB. Table2. GA operator parameters Membershp functons optmzaton Populaton sze Selecton type Elte count Crossover type Crossover rate Mutaton type Mutaton rate Stoppng generaton 4 Tournament 2 Scattered.85 Unform. Rule base optmzaton Populaton sze Selecton type Elte count Crossover type Crossover rate Mutaton type Mutaton rate Stoppng generaton 5 Tournament 25 Scattered.85 Unform 3 STORY7-DISPLACEMENT-ELCENTRO 5 STORY7-DISPLACEMENT-KOBE 2 LQR FUZZY-NET GA-FLC LQR Fgure8. Comparson of controlled roof dsplacement of the example frame calculated by the GA-FLC system wth the LQR controller 2 Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25

9 FORCE (N) Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated.5 2 x 4 STORY7-FORCE-ELCENTRO GA-FLC LQR Fgure9. Comparson of roof control force of the example frame calculated by the GA-FLC system wth the LQR controller The network was constructed based on the nformaton of sesmc structural response aganst an earthquake exctaton whch obtaned from several dynamc tme hstory analyss of a structure. Capablty to learn from data and generalzaton power of ts result on a set of test patterns enables the neural network to become a strong predctng and classfcaton tool. In ths study, a feed forward neural network wth an adaptve back propagaton tranng method was constructed and traned to generate predcton of reactve velocty and dsplacement at the current state from prevous measurements. In the back propagaton tranng, the learnng rate s determned by ensurng the decrease of the error functon of the nput-output tranng patterns at each tranng cycle. The neural network conssts of an nput layer that conssts of 22 neurons, a mddle hdden layer ncludng 4 neurons and an output layer wth 4 neurons as shown n Fgure. Fgure. Archtecture of the neural network The sgmod functon and lnear functon are chosen as the actvaton functon of hdden and output layer, respectvely. In order to predct the dynamc responses of the structure accurately, the most drectly and mportant factors affectng the predcted dynamc responses are consdered. Therefore, nputs to ths network are the delayed outputs (velocty and dsplacement), the delayed control forces, and the delayed earthquake nputs (acceleraton). The structure of nput and output tranng vectors s shown n Fgure. In order to ncrease ablty of tranng and generalzaton of network, nput and output data were normalzed by scalng n [-, ]. As the nputs are appled to the neural network, the dfference between the network outputs and the measured responses s processed back through the network to update the weghts and bases of the neural network untl the outputs match closer the measured responses. Ths process contnues for all the tranng data set untl the obectve error s satsfed. In ths paper, the sum square error functon s adopted to assess the network. Sesmc Response Mtgaton of GA-ANN-FLC Controller In order to evaluate the contrbuton of the ncorporaton of the neural network response predcton and nvestgate the effcency of the proposed control strategy to enhance tme delay effect, the control Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25 2

10 FORCE (N) DISP (cm) Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated process s establshed by consderng tme delay of.6 seconds. Fgure llustrates nfluence of the tme delay under Elcentro earthquake. When tme delay happens, GA-FLC system s not able to suppress structural response approprately. It can be seen that the neural network s able to estmate the dynamc responses of the structure as nputs to the fuzzy controller to determne control force of FDSAB and GA-ANN-FLC control scheme greatly reduces the roof dsplacement when neural network s ntegrated n the control strategy. The requred control force of the GA-FLC system s also compared to requred control force of GA-ANN-FLC strategy wth tme delay of.6 seconds n Fgure 2 for dsplacement mtgaton of 7th floor subected to Elcentro exctaton. As can be observed, ntegratng the neural network also results n reducng control force and GA-ANN-FLC system s able to solve tme delay problem. However, as dscussed prevously, LQR algorthm s superor n reducng control force. 5 STORY7-DISPLACEMENT-ELCENTRO GA-FLC GA-ANN-FLC Fgure. Comparson of controlled roof dsplacement of the example frame calculated by the GA-ANN-FLC system wth the GA-FLC controller for the tme delay of T =.6 seconds. 3 x 4 2 GA-FLC GA-ANN-FLC STORY7-FORCE-ELCENTRO Fgure2. Comparson of roof control force of the example frame calculated by the GA-ANN-FLC system wth the GA-FLC controller for the tme delay of T =.6 seconds CONCLUSION Ths study nvestgates performance of a GA-desgned FLC (GA-FLC) for vbraton control of structures equpped wth frcton dampng system wth amplfyng braces (FDSAB). In the desgn of the FLC system the velocty and dsplacement of stores are consdered as the feedback to the FLC. The genetc algorthm s used to optmze the parameters of membershp functons and fnd approprate fuzzy control rules n order to gan the maxmum dsplacement mtgaton. The proposed control strategy (GA-FLC) s mplemented for control of a seven story shear frame confgured wth FDSAB. For comparson purposes, LQR controller system s also desgned to control the buldng response. Based on numercal study, whle GA-FLC controller requres more control force, t s more effectve than LQR method n dsplacement response reducton. In the case of tme delay problem, a neural network s employed n control strategy as a state estmator (GA-ANN-FLC) and the control forces of the FDSAB are determned quckly by the fuzzy controller. The dsplacement response of controlled frame sgnfcantly decreases, when neural network s ntegrated wth the genetc-fuzzy controller. GA-ANN-FLC has capablty to enhance tme delay effect and approprately controls the dsplacement response of the frame. Based on numercal modellng of the example frame, the proposed methodology can be used to control of FDSAB equpped systems n an effectve way to decrease structural vbratons. 22 Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25

11 Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated REFERENCES [] Gaul L, Hurlebaus S, Wrntzer J, Albrecht H. "Enhanced dampng of lghtweght structures by sem- actve onts. " Acta Mech 95: [2] Dyke SJ, Spencer BF Jr, San MK, Carlson JD. Modelng and control of magnetorheologcal dampers for sesmc response reducton. Smart Mater Struc 5: [3] Xu YL, Qu WL, Ko JM. Sesmc response control of frame structures usng magnetorheologcal/ electrorheologcal dampers. Earthq Eng Struc Dyn 29: [4] Y F, Dyke SJ, Cacedo JM, Carlson JD. Expermental verfcaton of mult-nput sesmc control strateges for smart dampers. J Eng Mech 27(): [5] Zhang J, Roschke PN. Actve control of a tall structure excted by wnd. J Wnd Eng Ind Aerodyn 83: [6] McClamroch NH, Gavn HP. Closed loop structural control usng electrorheologcal dampers. In Proceedngs of the Amercan control conference, Seattle [7] Hemenz GJ, Cho Y-T, Wereley. Sesmc control of cvl structures utlzng sem-actve MR bracng systems. In Proceedngs of smart systems for brdges, structures, and hghways conference, Newport Beach. 2. [8] Jansen LM, Dyke SJ. Semactve control strateges for MR dampers: comparatve study. Journal of Engneerng Mechancs (ASCE) 26(8): [9] Paraskevopoulos PN. Modern Control Engneerng. Marcel Dekker, Inc. 22. [] N YQ, Chen Y, Ko JM, Cao DQ. Neuro-control of cable vbraton usng sem-actve magnetorheologcal dampers. Eng Struc 24: [] Zhou L, Chang CC, Spencer BF Jr. Intellgent technology-based control of moton and vbraton usng MR dampers. Earthq Eng Eng Vb ():. 22. [2] Xu ZD, Shen YP, Guo YQ. Sem-actve control of structures ncorporated wth magnetorheologcal dampers usng neural-networks. Smart Mater Struc 2: [3] Cho KM, Cho SW, Jung HJ, Lee IW. Sem-actve fuzzy control for sesmc response reducton usng magnetorheologcal dampers. Earthq Eng Struc Dyn 33: [4] Wlson CMD, Abdullah MM. Structural vbraton reducton usng fuzzy control of magnetorheologcal dampers. ASCE Structures Congress, New York. 25. [5] Ok SY, Km DS, Park KS, Koh HM. Sem-actve fuzzy control of cable-stayed brdges usng magneto-rheologcal dampers. Engneerng Structures 29: [6] Das D, Datta TK, Madan A. Semactve fuzzy control of the sesmc response of buldng frames wth MR dampers. Earthquake Engng Struct. Dyn 2. DOI:.2/eqe.2. [7] Kma HS, Roschke PN. Desgn of fuzzy logc controller for smart base solaton system usng genetc algorthm. Engneerng Structures 28: [8] Bdokht K.K., Moharram H. and Fayez A. Sem-actve fuzzy control for sesmc response reducton of.buldng frames usng SHD dampers. Structral Control and Health Montorng. 9(3), [9] Btaraf M, Ozbulut OE, Hurlebaus S, Barroso L. Applcaton of sem-actve control strateges for sesmc protecton of buldngs wth MR dampers. Engneerng Structures 32: [2] YK Wen J, Ghabus P, Nkzad VK. Control of Structures Usng Neural Networks. Smart Master structures 4, [2] Zhang J, Roschke PN. Neural network smulaton of magnetorheologcal damper behavor. Internatonal Conference on Vbraton Engneerng, Dalan, Chna, [22] Huang SJ, Lan RJ. A combnaton of fuzzy logc and neural network controller for multplenput multple-output systems. Int J Sys Sc 3(3): [23] ZhaoDong X, YngQng G. Integrated ntellgent control analyss on sem-actve structures by usng magnetorheologcal dampers. Scence n Chna Seres E: Technologcal Scences 5: [24] Gluck J, Rbakov Y. Sem-actve Frcton System wth Amplfyng Braces for Control of MDOF Structures. The structural desgn of tall buldngs, N., Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25 23

12 Hosen Ghaffarzadeh Sem-Actve Control of Structures Equpped wth FDSAB Usng Integrated [25] Rbakov, Y., Gluck, J. Actve Controlled Frcton Damped MDOF Structures wth Varable Stffness 8th Canadan Conference on Earthquake Eng., Vancouver, pp [26] Nshtan A, Ntta Y, Ishbash Y. Structural control based on sem-actve varable frcton dampers. Proceedngs of the 2th World Conference on Earthquake Engneerng, New Zealand. [27] Mualla IH, Belev B. Performance of steel frames wth a new frcton damper devce under earthquake exctaton. Eng Struct 24(3): AUTHOR S BIOGRAPHY Hosen Ghaffarzadeh receved hs PhD n Structural Engneerng 25 from Shraz Unversty, Iran. Hs research nterests are Structural Dynamcs, Structural Control and Damage detecton. He s currently an assocate professor n the Department of Cvl Engneerng at the Unversty of Tabrz, Tabrz, Iran. 24 Internatonal Journal of Emergng Engneerng Research and Technology V3 I3 March 25