Jtam-A4.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 51, 2, pp. 275-286, Warsaw 2013 ACTIVE VIBRATION CONTROL OF SMART COMPOSITE BEAMS USING PSO-OPTIMIZED SELF-TUNING FUZZY LOGIC CONTROLLER Nemanja Zorić, Aleksandar Simonović, Zoran Mitrović, Slobodan Stupar University of Belgrade, Faculty of Mechanical Engineering, Serbia e-mail: nzoric@mas.bg.ac.rs This paper presents the optimized fuzzy logic controller (FLC)with on-line tuning of scaling factors for vibration control of thin-walled composite beams. In order to improve the per- formances and robustness properties of FLC, the proposedmethod adjusts the input scaling factors via peak observer. The membership functions of the proposed FLC are optimized using the particle swarmoptimization (PSO) algorithm.The composite beam ismodeled by the third-order shear deformation theory (TSDT) anddiscretized byusing the finite element method. Several numerical examples are provided for the cantilever composite beam under a periodic excitation and periodic excitation with an unexpected disturbance. In order to present the efficiency of the proposed controller, the obtained results are comparedwith the corresponding results in the cases of the optimized FLCs with constant scaling factors and LQR optimal control strategy. Key words: optimal vibration control, fuzzy logic control, smart beams 1. Introduction Thin-walled composite structures are used for aerospace applications such as aircraft wings, wind turbine blades and helicopter blades. The appearance of unwanted vibrationmay result in instability of the system and, also, can reduce structural life and lead to catastrophic failure. In order to control vibrations of thin-walled structures during operation, conventional structures have been combined with sensing and actuating mechanisms to develop “smart structures” (Gabbert, 2002). In recent years, a great number of researches have been using piezoelectric materials as distributed actuators and sensors (A/Ss) for active vibration control. The dynamic performance and functionality of smart structures depend on the control algo- rithm. There are numerous control algorithms which can be applied for vibration suppression of thin-walled structures with distributed piezoelectric A/Ss. The most commonly used con- trol algorithms are “classical” control algorithm such as direct proportional feedback, constant gain velocity feedback (CGVF) and constant amplitude velocity feedback (CAVF) control (Ku- mar and Narayanan, 2008; Kapuira and Yasin, 2010). Optimal control algorithms (LQR and LQG) for active vibration suppressionwere adopted byKumar andNarayanan (2008), Roy and Chakraborty (2009), Kapuira and Yasin (2010) and Gabbert et al. (2005). These control algorithms can provide adequate control managment for wide classes of pro- blems related to active vibration suppression, but they are sensitive to structure characteristics and recquire an exact mathematical model of a structure even for the collocated system. Also, they are sensitive to operating conditions, furthermore, it is difficult to adjust controller gains. An alternative is the use of intelligent control algorithms based on soft computing schemes such as fuzzy logic control (FLC) algorithms. The fuzzy set theory was established by Zadeh (1965) and it has been extensively researched in various fields of engineering. The main advantage of FLC over conventional control approaches is that the FLC is considered artificial intelligence where control laws are designd by human intelligence based on expert’s experience, not by a 276 N. Zorić et al. deterministic numerical calculation. FLC does not recquire the accurate mathematical model of the controlled object and it can represent almost any deterministic controller. Therefore, FLC method has been applied widely for active vibration control of flexible structures (Sharma et al., 2007; Marinaki et al., 2010; Wei et al., 2010). Fuzzy control design is composed of three important stages. The first stage is the inference rules design, the second is tuning of membership functions, and the third is tuning of control parameters. This paper deals with active vibration control of smart composite beams using an optimized self-tuning fuzzy logic controller (FLC). Input variables in FLC are scaled with scaling factors, and, in order to improve theperformanceand robustnessproperties ofFLC, these scaling factors are adjusted via peak observer. Themembership functions of the proposed FLC are optimized using the particle swarm optimization (PSO) algorithm. The composite beam is modeled by the third-order shear deformation theory (TSDT) and discretized using the finite element method. Numerical examples are provided for the cantilever composite beam under a periodic excitation and a periodic excitation with unexpected disturbance. The results obtained using the proposed self-tuning FLC are then compared with the corresponding results in the cases of PSO-optimized FLC with constant scaling factors and LQR optimal control strategy. 2. Governing equations 2.1. Coupled equations of motion The laminated composite beamwith integrated piezoelectric actuators and sensors conside- red here is presented in Fig. 1. Both elastic and piezoelectric layers are supposed to be thin, so a plane stress state can be assumed. The sensors and actuators are perfectly bonded on the upper and lower surfaces at different locations along the length of the beam. It is assumed that they span the entire width of beam. The elastic layers are obtained by setting their piezoelec- tric coefficients to zero. A single mechanical displacement field is considered for all layers while electric displacements are considered for each layer independently. Fig. 1. Composite beamwith integrated piezoelectric layers The beam is discretized using the finite element method based on the third-order shear deformation theory (Heyliger and Reddy, 1985). The formulation results in a coupled finite ele- mentmodel withmechanical (displacements) and electrical (potentials of piezoelectric patches) degrees of freedom (Zorić et al., 2012) Mü+Cdu̇+K ∗ u=Fm− (Kme)AφAA (2.1) where u presents the vector of generalized mechanical displacmenets, M presents the mass matrix, (Kme)A is the piezoelectric stiffness matrix of the actuator, Cd is the dampingmatrix, φAA is the vector of external applied voltage on the actuators, Fm is the vector of external forces and K∗ is the coupled stiffness matrix given as K ∗ =Km+(Kme)A(Ke) −1 A (Kme) T A+(Kme)S(Ke) −1 S (Kme) T S (2.2) Active vibration control of smart composite beams... 277 where Km presents the elastic stiffness matrix, (Kme)S is the piezoelectric stiffness matrix of the sensor and (Ke)A and (Ke)S are the dielectric stiffnessmatrices of the actuator and sensor, respectively. 2.2. Modal analysis For practical implementation, the obtainedmodel needs to be truncated,where only the first few modes are taken into account. Thus, the displacement vector can be approximated by the modal superposition of the first r modes as u≈Ψη (2.3) where Ψ presents the modal matrix, and η the vector of modal coordinates. Using Equation (2.3), Equation (2.1) can be transformed in the reducedmodal space as η̈+Λη̇+ω2η=ΨTFm−Ψ T(Kme)AφAA (2.4) where ω2 presents the diagonal matrix of the squares of the natural frequencies, and Λ= diag i=1,r(2ζiωi) (2.5) presents the modal dampingmatrix in which ζi is the natural modal damping ratio of the i-th mode. 2.3. State-space representation Equation (2.4) can be expressed in the state-space form as Ẋ=AX+BφAA+d (2.6) where X= { η η̇ } A= [ 0 I −ω2 −Λ ] B= [ 0 B ] = [ 0 −ΨT(Kme)A ] d= [ 0 ΨTFm ] (2.7) present the state vector, the systemmatrix, the controlmatrix disturbance respectively, where I and 0 are the appropriately dimensioned identity and zero matrix. The sensor output equation can be written as YS =CSX (2.8) where CS presents the outputmatrix which depends on themodalmatrix and sensor piezoelec- tric stifeness matrix. 3. Fuzzy logic control 3.1. Design of the fuzzy logic controller The idea of fuzzy logic control (FLC) is using linguistic directions as a basis for control. By incorporating human expertise into the fuzzy IF-THEN rules, FLC can be embedded into a 278 N. Zorić et al. closed-loop system, similarly to conventional controllers. Themost used fuzzy control inference is theMamdani fuzzy inference, whose i-th rule can be written as follows Ri : if I1 =Ai ∧ I2 =Bi then O=C (3.1) where I1 and I2 present the input variables, O is the output variable, and Ai, Bi and Ci are the linguistic values of the fuzzy variables. In general, FLC consists of four principal elements: fuzzification, rule base, decision making and defuzzification. The first stage in building the fuzzy controller is choosing input/output parameters. Two commonly input variables are the error and the error derivative. In this study, the inputs are modal displacement η and its time derivative-modal velocity η̇. The output variable is control voltage applied on the actuator φAA. The fuzzy logic controller uses crisp data directly from the sensor. These are converted into linguistic values through the fuzzification process. Each input and output fuzzy variable is defined in the fuzzy space in the form of five linguistic variables namelyNB (negative big), NS (negative small), ZE (zero), PS (positive small) and PB (positive big). In this study, the trapezoid membership function is used to present NB and PB variables and the triangular membership function to present NS, ZE and PB variables. The universes of discourse of the inputs and the output are set to be [−1,1], hence, the inputs need to be scaled, thus its minimum value be −1, andmaximum 1. The scaling of the inputs is performed on the following way E =Kdη EC =Kvη̇ (3.2) where E and EC present the error and the error derivative in the fuzzy set, and Kd and Kv present the displacement and the velocity scaling factor, respectively. Also, the output from the fuzzy set needs to be scaled in the following way φAA =KactU (3.3) where U presents the output from the fuzzy set, and Kact is the output scaling factor which is equal to themaximumallowable voltage which can be applied to the piezoelectric actuator. The membership functions of the inputs and the output are presented in Fig. 2. The membership functions are parameterized by parameters presented by the matrix α, which are depicted in Fig. 2. Fig. 2. Membership functions of the inputs and the output Observing Equation (3.1), and considering the number of fuzzy linguistic variables for each input and output, it can be concluded that the number of fuzzy inference rules is 25. They are presented in amarix form shown in Table 1. For example, the rule described by the second row and second column in Table 1 reads: “IF E is NS and EC is NS, THENU is PS”. The results of fuzzy inference have to be transformed into a numerical output value through the process of defuzzification. In this study, the center average (centroid) defuzzificationmethod is used. Determination of parameters of the membership functions requires expert knowledge. No fixed process for determination of these parameters exists. In this study, these parameters are optimized in order to maximize control performances. Active vibration control of smart composite beams... 279 Table 1. Inference rules used in the proposed FLC EC NB NS ZE PS PB NB PB PB PB PS PS NS PS PS PS ZE ZE E ZE PS PS ZE NS NS PS ZE ZE NS NS NS PB NS NS NB NB NB 3.2. Self-tuning fuzzy logic controller Since the universes of discourse of the input variables are in the range [−1,1], the scaling factors Kd and Kv have to be chosen in such a way that they transform the input variables from the sensor to the fuzzy controller to be in the range [−1,1]. According to that, the scaling factors for modal displacement andmodal velocity can be calculated as follows Kd = 1 |ηmax| Kv = 1 |η̇max| (3.4) where ηmax and η̇max present the amplitudes of modal displacement and modal velocity. The amplitude is a time varying value and, also, the external excitations which cause vibration have a stochastic nature. Because of that, it is difficult to determine these factors off-line, and keeping them constant through active vibration suppression leads to a decrease of control performances (Wilson, 2005). Qiao and Mizumoto (1996) proposed the tuning of the parameters using peak observer for PID-type FLC. The peak observer, proposed by Qiao andMizumoto (1996), keeps watching the system output and transmits a signal at each peak, and measures the apsolute peak. In this study, this approach is adapted to the problems of vibration reduction, so, the peak observer is constructed for each input variable in the fuzzy controller. Beside that, the peak observer is modified and, beside watching the peaks, the presented observer also monitors the increase in the ampliude. A block diagram of the self-tuning FLC is presented in Fig. 3. Fig. 3. Block diagram of self-tuning FLC The presented peak observer monitors the input, and calulates the rates of the input. Con- sidering the k-th sampling time, the current and previous rates of the modal displacement are ∆η(k) = η(k)−η(k−1) ∆η(k−1)= η(k−1)−η(k−2) (3.5) and thedisplacement scaling factor is tuned in theparameter regulator according to the following expression 280 N. Zorić et al. Kd(k)=              1 |η(k−1)| for ∆η(k)∆η(k−1)¬ 0 1 |η(k)| for ∆η(k)∆η(k−1)> 0 and |η(k)|> 1 Kd(k−1) Kd(k−1) other (3.6) For tuning of the velocity scaling factor, the same procedure as for tuning of the displacement scaling factor can be applied. Figure 4 illustratively presents how the scaling factors are tuned via the peak observer. Fig. 4. Tuning of the displacement scaling factor via peak observer ConsideringFig. 4a, it canbe seen that ∆η(k−1)> 0and ∆η(k)< 0, thus, η(k−1)η(k)< 0, and according to Equation (3.6), the displacement scaling facor is scaled according to this Equation. On the other hand, considering the (k + 1)-th sampling time, it is obvious that ∆η(k+1) < 0, and η(k− 1)η(k) > 0, so the displacement scaling factor, in the (k+1)-th sampling time, does not change its current value. During operation, a structure can be affected by various disturbances, so, amplitude will be increased. In this case, the scaling factors have to be tuned.This is givenby the second row inEquation (3.6) and illustrated inFig. 4b.Considering Fig. 4b, it can be seen that ∆η(k−1)> 0 and ∆η(k)> 0, so their product is a positive value, but the absolute value of amplitude in the current, k-th, sampling time η(k) is larger than the absolute value of the last peak η(k−2). In this case, the displacement scaling factor needs to be tuned so that the input in FLC be in the range [−1,1]. 3.3. Optimization criteria The next step in designing of FLC is the determination of parameters of the membership functions. In this study, themembership functions are parameterizedwith parameters presented by the matrix α. The aim is to find optimal values of these parameters such that the active vibration suppression is improved. In order to improve the control performances of FLC, the optimization problem can be written as follows maxJ = 1 ‖η‖ (3.7) where J presents the objective functionwhich has to bemaximized, and ‖η‖ presents L2 norm of the vector of modal displacement. Constraints of this optimization problems are Subject to 0<αi,j ¬ 1 i=1, . . . ,3 j=1,2 αi,2 <αi,1 i=1, . . . ,3 (3.8) Active vibration control of smart composite beams... 281 Incorporating constraints into the optimization problem, the objective function can be transfor- med as J =      1 ‖η‖ if constraints are not violated 0 if constraints are violated (3.9) 4. Optimization implementation using particle swarm optimization technique The particle swarm optimization (PSO) has been inspired by the social behavior of animals such as fish schooling, insect swarming and birds flocking. It was introduced by Kennedy and Everhart (1995). The system is initialized with a population of random solutions (cal- led particles). Each particle represents a potential solution of the problem, and it is treated as a point in a m-dimensional space. For a given i-th particle, its position is represented as (pi) = (pi1,pi2, . . . ,pid, . . . ,pi,m) in which every coordinate presents a parameter which has to be optimized, and m presents the number of these parameters. The current position of every particle is affected by three factors: its own velocity: (vi) = (vi1,vi2, . . . ,vid, . . . ,vi,m), the best position it has achieved (best local position) which is determined by the highest value of the objective function encountered by this particle in all its previous iteration and overall best posi- tion achieved by all particles (best global position), which is determined by the highest value of the objective function encountered in all the previous iteration. The particle changes its velocity and position in the following way vk+1 id =χvkid+c1r1(lid−p k id)+c2r2(gd−p k id) pk+1 id = pkid+v k+1 id i=1, . . . ,n d=1, . . . ,m (4.1) where χ is the inertia weight, c1 is the cognition factor, c2 is the social learning factor, r1 and r2 are random numbers between 0 and 1, the superscript k denotes the iterative generation, n is the population size, lid and gd are the best local and the best global position. The cognition and social learning factors are usually set as c1 = c2 = 1.5. In this study, the coordinates of particles are the parameters given by the matrix α. 5. Numerical studies In this example, a cantilever symmetric laminated beam is considered. The length of the beam is 0.5m, and its width is 0.025m. The beam is made of eight graphite-epoxy (carbon-fiber reinforced) layers. The thickness of each layer is 0.25mmand orientations are (90◦/0◦/90◦/0◦)S. The piezoelectric actuator and sensor are made of PZT. Their thicknesses are 0.2mm, and lengths are 50mm. The actuator and sensor are placed at the root of the beam and they are collocated.Theallowable electric field of piezoceramicmaterials is around500-1000V/mm.Since the thickness of the actuator is 0.2mm, themaximumallowable voltage has been taken as 200V. Material properties of the graphite-epoxy layer and PZT are given in Table 2. The beam is discretized with 50 finite elements, and it is subjected to a periodic loading of 0.02sin(20t)N at the tip. Only the firstmode is considered. The number of randomly generated particles is 100, and number of iteration is 100. The inertia weight is lineary varied from 1 to 0.5 through iterations. The cognition and social learning factors in the PSO algorithm are set as c1 = c2 = 1.5. The initial values of displacement and velocity scaling factors for self-tuning FLC are: Kd = 10000 and Kv = 10. During simulation, the sampling time is set to be 1ms, and only the first mode is considered. The natural modal damping ratio is set to be 0.2%. The 282 N. Zorić et al. obtained results for self-tuningFLCafter optimization is presented inTable 3. Figure 5 presents the obtained membership functions. The tip displacement history is depicted in Fig. 6. Table 2.Material properties of graphite-epoxy and PZT Material properties Graphite-Epoxy PZT Modulus of elasticity, Y1 [GPa] 174 63 Modulus of elasticity, Y2 [GPa] 10.3 63 Shear modulus,G13 [GPa] 7.17 24.6 Shear modulus,G23 [GPa] 6.21 24.6 Poisson’s ratio, ν12 0.25 0.28 Density, ρ [kg/m3] 1389.23 7600 Piezoelectric constant, e31 [C/m 2] – 10.62 Permittivity constant, k33 [F/m] – 15.55 ·10 −9 Table 3. Initial displacement and velocity scaling factors, parametermatrix, objective function value andmaximim applied voltage for self-tuning FLC obtained by PSO Max. Kd Kv α J actuator voltage [V] 10000 10    0.056 0.028 0.06 0.03 0.92 0.46    1.11 ·104 164.5 Fig. 5. Membership functions for the inputs and output of the self-tuning FLC optimized by PSO Fig. 6. Tip displacement history: (a) self-tuning FLC optimized by PSO, (b) comparison of self-tuning FLC optimized by PSOwith uncontrolled tip displacement In order to present the influence of initial settings of the scaling factors Kd and Kv and performances of the controller optimized for the specific initial scaling factors with different initial factors used, the numerical simulation is performed with the following initial values: Active vibration control of smart composite beams... 283 Kd =0,Kv =0 and Kd =1000, Kv =10. Tip displacement histories for different initial values of the factors are depicted in Fig. 7a. From Fig. 7a, it can be concluded that different initial values of the scaling factors affect only the first 0.3s of vibration suppression. After that, these factors are adjusted, and the control performances do not differ from the initial values of scaling factors which were used for optimization. Fig. 7. Tip displacement histories: (a) for different initial values of scaling factors, (b) comparison self-tuning FLCwith FLCwith constant factors and LQR optimal control The next aim is the comparison of the presented optimized self-tuning FLCwith the optimi- zed FLC having constant scaling factors and optimal LQR. For this purpose, the membership functions of FLCwith constant factors are optimized with the proposed optimization technique for two examples considering the same loading as in the self-tuningFLC. In thefirst example, the scalling factors are set to be Kd =10000 and Kv =10, and in the second example: Kd =1000 and Kv = 10. The obtained results are presented in Table 4. For optimal LQR control, the weighting matrices Q and R are obtained by a trial and error solution, and the reuslts are presented in Table 5. The tip deflection histories are depicted in Fig. 7b. Table 4.Displacement and velocity scaling factors, parameter matrix, objective function value andmaximim applied voltage for FLCwith constant scaling factors obtained by PSO Max. Kd Kv α J actuator voltage [V] First example 10000 10    0.05 0.025 0.86 0.43 0.764 0.382    9.11 ·103 155.5155 Second example 1000 10    0.056 0.028 0.06 0.03 0.92 0.46    1.53 ·103 97.151 Table 5.Matrices Q and R andmaximum control voltage for LQR optimal control Max. Q R actuator voltage [V] 1012I2×2 1 170 284 N. Zorić et al. From Fig. 7b, it can be concluded that the proposed self-tuning FLC is more efficient in active vibration suppression than the classic FLCwith constant scaling factors andLQRoptimal control. Comparing themaximumvoltage obtained in the self-tuning FLC (see Table 3) and the maximum voltage obtained in LQR (see Table 5), it is evident that, beside better efficiency in active vibration suppression, the self-tuningFLCprovides a little lessmaximumapplied voltage than LQR. Furthermore, in order to investigate the robustness of the self-tuning FLC optimized PSO and its ability to work for different loadings, the obtained parameters for a specific loading is used for differential loadings. In other words, the parameters obtained by using PSO for loading 0.02sin(20t), are applied when the loadings are equal 0.02sin(10t), 0.02sin(30t) and 0.02sin(40t). The tip displacement histories for different loadings are presented in Fig. 8. From Fig. 8, it can be concluded that the proposed controller, that is optimized for one loading, can be applied successfully when other loadings are used. Fig. 8. Tip displacement histories of self-tuning FLC for different loadings The next study about robustness of the proposed FLC is carried out under a periodic exci- tation with unexpected disturbance. The tip of the beam is subjected to the periodic loading of 0.02sin(20t)N, and in 0.5s, the unexpected disturbance of 1Nwith duration of 1ms is applied. The response of the self-tuning FLC is compared with the responses of FLCs with constant scaling factors and LQR optimal control. Fig. 9. Tip displacement histories for a harmonic loading with unexpected disturbance The tip displacement histories are plotted in Fig. 9. Histories of the applied control voltages for the self-tuning FLC and LQR are depicted in Fig. 10. Observing Fig. 9, it can be concluded that the performances of the self-tuning FLC and of LQR are not significantly affected. For the Active vibration control of smart composite beams... 285 first example of theFLCwith constant factors (Kd =10000,Kv =10), it canbe seen formFig. 9 that, after appearing of the unexpected disturbance, the amplitude is significantly increased.On the other hand, comparing the applied control voltages during active vibration suppression for the self-tuning FLC and LQR, given in Fig. 10, it is evident that in the case of LQR, the maximum applied voltage is drastically increased, near 2000V. In this case, depolarization of the piezoelectric actuator is inevitable. Fig. 10. Histories of the applied control voltage; (a) self-tuning FLC, (b) LQR 6. Conclusions This paper presents active vibration control of smart composite beams using PSO-optimized self-tuningFLC. Input variables in FLC are scaledwith scaling factors, and, in order to improve theperformanceand robustness ofFLC, these scaling factors are adjustedviapeakobserver.The membership functions of the proposedFLCare optimized using theParticle swarmoptimization (PSO) algorithm. Taking intoaccount several numerical examplesperformed for thecantilever compositebeam, where thePSO-optimized self-tuningFLC is comparedwithPSO-optimizedFLCswith constant factors and the LQR control algorithm, one finds that the PSO-optimized self-tuning FLC is muchmore effective in vibration control than other investigated control algorithms. Although the proposed optimized self-tuning FLC is applied to the beam in the case of the first mode response, taking into consideration all advantages presented here, this technique can be applied to more complex structures like plates and shells, and for multimodal responses. Acknowledgment This work is supported by theMinistry of Science andTechnologicalDevelopment of the Republic of Serbia through Technological Development Projects No. 35035 and No. 35006. References 1. Gabbert U., 2002, Research activities in smart materials and structures and expectations to future developments, Journal of Theoretical and Applied Mechanics, 40, 3, 549-574 2. GabbertU.,Nestorović-TrajkovT.,KöppeH., 2005,Controllerdesign and implementation for active vibration suppression of a piezoelectric smart shell structure, Journal of Theoretical and Applied Mechanics, 43, 3, 487-500 3. Heyliger N.D., Reddy N., 1985,A higher order beamfinite elements for bending and vibration problem, Journal of Sound and Vibration, 126, 309-326 286 N. Zorić et al. 4. Kaipura S., Yasin M.Y., 2010, Active vibration suppression of multilayered plates integrated with piezoelectric fiber reinforced composites using an efficient finite element model, Journal of Sound and Vibration, 329, 3247-3265 5. Kennedy J., Everhart R.C., 1995, Particle swarm optimization, Proceedings of the IEEE In- ternational Conference on Neural Networks, 4, 1942-1948 6. Kumar R.K., Narayanan S., 2008, Active vibration control of beams with optimal placement of sensor/actuator pairs, Smart Materials and Structures, 17 7. Marinaki M., Marinakis M., Stavroulakis G.E., 2010, Fuzzy control optimized by PSO for vibration suppression of beams,Control Engineering Practice, 18, 618-629 8. Qiao W.Z., Mizumoto M., 1996, PID tupe fuzzy controller and parameters adaptive method, Fuzzy Sets and Systems, 78, 1, 23-35 9. Roy T., Chakraborty D., 2009, Optimal vibration control of smart fiber reinforced composite shell structures using improved genetic algorithm, Journal of Sound and Vibration, 319, 15-40 10. Sharma M., Singh S.P., Sachdeva B.L., 2007, Modal control of a plate using a fuzzy logic controller, Smart Materials and Structures, 16, 1331-1341 11. Wei J., Qiu Z., Han J., Wang Y., 2010, Experimental comparison research on active vibra- tion control for flexible piezoelectric manipulator using fuzzy controller, Journal of Intelligent and Robotic Systems, 59, 31-56 12. Wilson C.M.D., 2005,Fuzzy Control of Magnetorheological Dampers for Vibration Reduction of Seismically Excited Structures, Ph.D.Thesis, Florida State University, Tallahassee 13. Zadeh L.A., 1965, Fuzzy sets, Information and Control, 8, 338-353 14. Zorić N., Simonović A.,Mitrović Z., Stupar S., 2012,Multi-objective fuzzy optimization of sizing and location of piezoelectric actuators and sensors,FME Transactions, 40, 1-9 Aktywna redukcja drgań „inteligentnych” belek kompozytowych za pomocą samostrojącego sterownika z logiką rozmytą i zoptymalizowanego metodą PSO Streszczenie Wpracy zaprezentowano zoptymalizowany sterownik z logiką rozmytą (FLC) o bieżąco dostrajanych współczynnikach skalowania zastosowany do redukcji drgań cienkościennych belek kompozytowych. Dla poprawy efektywności i stabilności pracy sterownika zaproponowano dostrajanie współczynników skalo- wania na wejściu poprzez śledzenie wartości szczytowej sygnału. Funkcje przynależności sterownika FLC zoptymalizowano algorytmem roju cząstek (PSO). Rozważaną belkę kompozytową opisanomodelem teo- rii odkształceń postaciowych trzeciego rzędu (TSDT) i zdyskretyzowanometodą elementów skończonych. Przedstawionokilkaprzykładówbelkiwspornikowejpoddanejwymuszeniuokresowemuznieoczekiwanym zakłóceniem tegowymuszenia.W celu zeprezentowania efektywności analizowanego sterownika uzyskane wyniki porównano z istniejącymi rezultatami odpowiadającymi sterownikom FLC o stałych współczyn- nikach skalowania i zoptymalizowanej strategii sterowaniawykorzystującej regulator liniowo-kwadratowy (LQR). Manuscript received April 11, 2012; accepted for print May 24, 2012