Jtam-A4.dvi JOURNAL OF THEORETICAL AND APPLIED MECHANICS 54, 4, pp. 1109-1123, Warsaw 2016 DOI: 10.15632/jtam-pl.54.4.1109 IMPLEMENTATION OF THE LQG CONTROLLER FOR A WIND TURBINE TOWER-NACELLEMODEL WITH AN MR TUNED VIBRATION ABSORBER Maciej Rosół AGH University of Science and Technology, Department of Automatics and Biomedical Engineering, Cracow, Poland e-mail: mr@agh.edu.pl Paweł Martynowicz AGH University of Science and Technology, Department of Process Control, Cracow, Poland e-mail: pmartyn@agh.edu.pl Vibration of a wind turbine tower is related to fatigue wear, influencing reliability of the whole structure. The current paper deals with the problem of Linear-Quadratic-Gaussian (LQG) tower vibration control using specially designed and built simulation and laboratory tower-nacelle models with a horizontally aligned, magnetorheological (MR) damper based tuned vibration absorber located at the nacelle. Force excitation applied horizontally to the tower itself, or to the nacelle, is considered. The MR damper LQG control algorithm, including the Kalman state observer and LQR (Linear-Quadratic-Regulator) controller is analysed numerically and implemented on the laboratory ground, in comparison with the systemwith a deactivated absorber. Simulation and experimental results are presented. Keywords:windturbine towervibration, tunedvibrationabsorber,MRdamper,LQGcontrol 1. Introduction Thewind turbines sector is an emerging one nowadays. Thewind load (and also sea waves load for offshore structures) that is varying in time as well as rotation of turbine elements are the major contributors to structural vibration of the tower and blades. Cyclic stress the tower is subjected to, may lead to a decrease in reliable operation time due to structure fatigue wear (Enevoldsen and Mork, 1996) or even failure accident. Tower vibration arises due to various excitation sources as variable wind/sea conditions and rotation of turbine elements (Jain, 2011). This vibration is generally lightly damped, especially considering low aeroelastic damping for the first tower lateral mode (Butt and Ishihara, 2012; Hansen et al., 2012; Matachowski and Martynowicz, 2012; Bak et al., 2012). The lateralmodes are excited due to theKarmanvortices, generator operation, sea waves variable load and rotatingmachinery unbalance rather than due to direct wind load variation/wind shear/differences in inflow conditions for each of the blades, and the blade passing effect, as for longitudinal modes. In the current research, tower vibration only is being analysed. The solutions utilised to reduce vibration of wind turbine towers include collective pitch control, generator electromagnetic torque control, and passive/semiactive/active tuned vibra- tion absorbers (TVAs) (Shan and Shan, 2012; Jelavić et al., 2007; Namik and Stol, 2011; Den Hartog, 1985; Oh and Ishihara, 2013; Tsouroukdissian et al., 2011; Rotea et al., 2010). TVAs arewidely spread structural vibration reduction solutions for slender structures. In the standard (passive) approach,TVAconsists of an additionalmovingmass, spring andviscous damperwho- se parameters are tuned to the selected (most often first) mode of the vibration (Den Hartog, 1985; Łatas and Martynowicz, 2012). Passive TVAs work well at the load conditions characte- rised with a single frequency to which they are tuned, but cannot adapt to a wider excitation 1110 M. Rosół, P. Martynowicz spectrum (Kirkegaard et al., 2002), thusmore advanced TVAs are implemented to change/tune TVAoperating frequency.Among them,magnetorheological (MR)TVAs are placed (Kirkegaard et al., 2002), as the use of MR damper (instead of viscous damper) guarantees a wide range of the resistance force, fast response times, low sensitivity to temperature changes and fluid conta- mination, high operational robustness, andminor energy requirements (Kciuk andMartynowicz, 2011; LordRheonetic, 2002; Sapiński andRosół, 2007, 2008; Snamina andSapiński, 2011; Sapiń- ski and Martynowicz, 2005) as compared to active systems. Simulations and experiments show that the implementation of the MR damper in the TVA system may lead to further vibration reduction in relation to the passive TVA (Martynowicz, 2014b, 2015, 2016; Koo andAhmadian, 2007). Within the scopeof the current research, therehavebeen specially developed andbuilt tower- -nacelle simulation and laboratorymodels inwhich all turbine components (nacelle, blades, hub, shaft, generator and possibly gearbox) are represented by nacelle concentrated mass and mass moments of inertia. Regarding variable geometric configuration of the structure resulting from changing rotor angular position, more detailed FEM analysis has been conducted using the full structuremodel, which demonstrated negligible influence of the rotor angle on tower structural dynamics (Matachowski and Martynowicz, 2012). Both simulation model and laboratory test rig of wind turbine tower-nacelle system give the possibility to model tower vibration under various aerodynamic, hydrodynamic, mechanical unbalance, changeable electromagnetic load, excitation sources (as mentioned before), etc. A horizontal concentrated force generated by the modal shaker may be applied either to the nacelle P(t) or to the tower itself at half of its height F(t). With the use of the MR damper, dedicated control solutions may be realised, in comparison to the system without TVA (i.e. TVA ‘locked’). Previous research showed the effectiveness of the ground hook control and its modification, sliding mode control, linear and nonlinear damping, adaptive control and open-loop system with various MR damper constant input current values (Martynowicz, 2014b, 2015, 2016). The Linear-Quadratic-Gaussian (LQG) control approach implementation is analysed within the scope of the current paper. The first bending mode of vibration is analysed here only, as higher modes reduction capabilities with MR TVA located at the nacelle are minor (Martynowicz, 2014b, 2015). Most of the applications of the LQGcontroller concern control of civil structures (buildings) excited by severe earth quakes or strong winds. The existing solutions of the LQG semiacti- ve control algorithm use, most frequently, the mathematical model of the analysed mechanical structure (Dyke et al., 1996a,b; Asai et al., 2013; Wang and Dyke, 2013). In opposition to the Linear-Quadratic-Regulator (LQR) algorithm, they do not need measurement of the full-state for all DOFs. TheKalman state observer is responsible for the estimation of unmeasurable state variables, based on the measured positions or accelerations. In many cases, obtaining a suffi- ciently accurate model is difficult, therefore some authors proposedmodel-free LQG semiactive control algorithms which do not need an accurate mathematical model (Asai and Spencer Jr., 2014), computing LQG parameters directly from the measurement data (Hjalmarsson et al., 1998; Kawamura, 1998; Favoreel, 1999). In the present paper, an output feedback strategy based on the measured position at a limited number of structure points is proposed. The LQG controller calculates the desired MR damper force on the basis of the state variables vector restored by the Kalman filter. The LQR problem is solved using a ’black box’ input-output linear model identified on the basis of the free vibration response of the tower-nacelle experimental model. Given themeasurements of the inputs and outputs of the unknown system, the matricesA,B,C andD of the estimate linear system are found. Themodel order of the ’black box’ is selected considering two state variables (position and velocity) for each selected structure point. The paper is organised as follows. In the next Section, the wind turbine tower-nacelle Com- sol/Simulink model with MR TVA is introduced. Then, the laboratory test rig is presented. Implementation of the LQG controller for a wind turbine... 1111 LQG controller synthesis including the Kalman filter is further described and followed byCom- sol/Simulink simulation and laboratory test rig experimental results. The paper is finishedwith several conclusions. 2. Wind turbine tower-nacelle model with MR TVA The beam modelling the tower is arranged vertically. The bottom end of the beam is fixed to the ground via additional foundation frame. A solid body modelling the nacelle is fixed to the upper end of the beam. TVA system incorporating the absorber mass, spring and MR damper is attached to the solid body representing the nacelle, and arranged horizontally. A diagram of the model, including the system of coordinates w-x, is shown in Fig. 1a. Fig. 1. (a) Diagram of the model, (b) The laboratory test rig Based on themodel assumptions andmathematical calculations results, aComsolMultiphy- sics finite element method (FEM) model of the tower-nacelle system was built as a ‘3D Euler Beam’ fixed at the bottom and free at the top, with an additional mass and mass moments of inertia defined at its top. A beam element (of length designated by l) with three nodes has been selected. The two edges are configured by applying material and cross-section parameters of the chosen tower material. The bottom node represents the tower-ground (tower-foundation) restrain, while two other nodes are ‘free’. The node at the tower midpoint (at x = x0 = l/2, see Fig. 1a), where deflection of the 2nd mode is close to maximal, is the ‘load point’, where the horizontal w-axis force F (F(t)) may be applied. The node at the top of the tower (at x= x1 = l) corresponds to the nacelle location, thus the mass and mass moments of inertia as well as concentrated load P (P(t)) are all assigned here. The FEM model assumes that angles are small and cross sections are perpendicular to the bending line (Euler-Bernoulli beam model). Also, the Rayleigh model (that is precise within a narrow frequency range only) of the tower material damping is assumed. These assumptions make the developedmodel to be adequate for small bending angles and within the 1st bending mode frequency neighbourhood only. Also, during the model identification process (subject to separate publication), correction (lowering) of Young’s modulus is necessary as the FEMEuler- Bernoulli beammodel is stiffer than the real structure due to neglecting shear deformation and additional restrains input by the finite/limited number of elements. 1112 M. Rosół, P. Martynowicz FEM Comsol Multiphysics model has been exported to MATLAB/Simulink with the ‘Ge- neral dynamic’ option. During exporting of the ‘Simulinkmodel’, forces F, and P are specified as inputs, while the tower tip (wx1/vx1) and tower midpoint (wx0/vx0) displacements/velocities along thew-axis are defined as output signals. After the exporting, FEM tower-nacelle model is available as aMATLAB structure, and the ComsolMultiphysics model is embedded into Simu- link diagram using ‘COMSOLMultiphysics Subsystem’ block. Thus, all 18 FEMmodel degrees of freedom are a part of the Simulink state vector (COMSOL, 2008). The MR TVA model is implemented as a standard Simulink diagram. TVA model with the hyperbolic tangent model of RD-1097-1 MR damper (Martynowicz, 2015) including linear bearing guides (see Section 3) friction force, LQG controller block with MR damper optimal (demanded) force output, incorporating the Kalman filter and LQR state feedback loop, linear guides friction force compensation by anMRdamper, andMRdamper inversemodel (to obtain demanded control current) are all embedded in the Simulink diagram. ‘COMSOLMultiphysics Subsystem’blockoutputsare fed to thedynamicsofMRTVAwithmassand stiffnessparameters (designated bymandk, respectively) tunedaccording toDenHartog (1985).Ageneral structure of the regardedmodel is shown in Fig. 2. It contains twomain blocks: the tower-nacelle system model and theMRTVAmodel. Its inputs are:F,P, whilewx0, vx0,wx1 and vx1 are the outputs. The forces produced by an MR damper, spring, and bearing guides friction are all added to a force excitation P, (Martynowicz, 2014b, 2015). Fig. 2. Structure of the simulationmodel; F , P – horizontal force applied at the point x0 and point x1, respectively,PMR –MRdamper force, iMR – current in theMR damper coil,wx0, vx0 – towermidpoint horizontal displacement and velocity, respectively,wx1, vx1 – tower tip (nacelle) horizontal displacement and velocity, respectively 3. Laboratory setup The analysedmodel has to fulfil various constraints imposed by the laboratory facility and pro- ject limitations, among others adequate dimensions, strength andmodalmasses of the structure, and mass of the absorber corresponding to the commercially available MR damper characteri- stics, to enable reduction of tower deflection amplitude for the nominalMRdamper stroke. It is assumed that at least partial dynamic similarity between the real-worldwind turbine (Vensys 82) tower-nacelle system and its scaled model has to be fulfilled, respecting the limited laboratory space and foundation permissible load (Martynowicz, 2014a; Snamina et al., 2014; Martynowicz and Szydło, 2013). Based on all of the assumptions and analyses, a Ti. Gr. 5 alloy rod has been selected tomodel the wind turbine tower, while Lord’s RD-1097-1 (Lord Rheonetic, 2002) MR damper has been used for TVA, andTMS2060E lightweight electrodynamic exciter (TMS, 2010) has been selected for excitation purposes. The parameters of TVA have been tuned for the 1st bending mode of vibration (Den Hartog, 1985). After several system reconfigurations, the absorber mass has been selected to be ca. 15% of the modal mass of the 1st bendingmode of the tower-nacelle model. Implementation of the LQG controller for a wind turbine... 1113 The detailed analysis of a similarity relation between the laboratory model and the full- -scale (Vensys 82) structure, including time and length similarity scale factors and determined geometrical andmaterial properties of the model, was presented by Snamina et al. (2014). The conducted partial dynamic similarity analysis ensures motion similarity of a selected pair of corresponding points (tower tips). The laboratory test rig of the wind turbine tower-nacelle withMRTVA system is presented in Fig. 1b. It is build according to the details specified above. It consists of a vertically oriented titanium alloy circular rod 1 (representing wind-turbine tower), and a system of steel plates 2 (representing the nacelle and turbine assemblies) fixed to the top of rod 1, with MR TVA embedded. The rod is rigidly mounted to steel foundation frame 3.MR TVA 4 is an additional massmoving horizontally along linear bearing guides, connectedwith the assembly representing thenacelle via the springandRD-1097-1MRdamper inparallel.RD-1097-1 damper,whose force depends on the current fed to its coil, is an actuator of such a vibration reduction system. The MRTVAoperates along the samedirectionas thevibration excitation applied to the system.The force generated by TMS 2060E exciter 5 may be applied either to rod 1 (modelling the tower) midpoint or to the system of steel plates 2 (modelling the nacelle/turbine) with the help of drive train assembly 6 of changeable leverage (enabling changeable force/displacement/velocity ranges). The excitation of the tower resulting fromblade rotation, rotatingmachinery unbalance as well as wind thrust on the rotor may be modelled by a concentrated load P applied to the nacelle/turbine,while thedirect (aerodynamic, includingbladepassingeffect, seawaves, ice, etc.) tower loadsmaybe reduced to the resultant concentrated forceF applied to the tower itself (e.g. at its midpoint). All themeasurements are gathered by PCwithMATLAB/Simulink/RT-CON based real-time environment that is also used for theMR damper control and excitation signal generation (Martynowicz, 2015, 2016). 4. LQG Controller synthesis The LQG (Linear-Quadratic-Gaussian) controllers are built for uncertain linear systems distur- bed by additive white Gaussian noises, having incomplete state information (Athans, 1971). The LQG is a combination of the Kalman filter with Linear-Quadratic Regulator (LQR). The separation principle allows each part of the LQG to be designed and tested independently. The LQG controller applies to both linear time-invariant and time-varying systems. It should be no- ted that the LQG control problem is one of the most fundamental problems of optimal control. Application of aKalman filter enables to restore unmeasured state variables and then use them in the LQR controller. A typical structure of the LQG regulator is shown in Fig. 3. Fig. 3. Structure of a LQG controller;u – control input of the process, ϑ – process noise (stochastic), ψ – measurement noise (stochastic), y – output of the process, ŷ – estimation of the process output, x̂ – estimation of the process state 1114 M. Rosół, P. Martynowicz The present paper concerns a discrete-time LQG control problem. The description of the LQG controller focuses on the following discrete-time linear system of equations, modelling the tower-nacelle withMRTVA system (block ‘Process’ in Fig. 3) x(k+1)=Ax(k)+Bu(k)+ϑ(k) y(k) =Cx(k)+Du(k)+ψ(k) (4.1) where x is the state vector, while the process and measurement noises, respectively: ϑ(k) and ψ(k) are independent, zeromean, white Gaussian random processes, satisfying (Qd,Rd – cova- riance matrices) E[ϑ(k)] =E[ψ(k)] = 0 E[ϑ(k)ϑT(k)] =Qd E[ψ(k)ψ T(k)] =Rd (4.2) The tower-nacelle model itself (‘Tower-nacelle model’ block in Fig. 2), as presented in Sec- tion 2, is linearizedwith regard to the following four state variables: tower tip (nacelle) horizontal displacementwx1 and velocity vx1, and tower midpoint horizontal displacementwx0 and veloci- ty vx0. All the other tower-nacellemodel state variables producedby theCOMSOLMultiphysics FEM model are ignored as not crucial concerning 1st (and even 2nd, regarding future applica- tions) bendingmode amplitudes detection andMRTVA control. TheLQGcontrol algorithmcanbe employed for semi-active control of the tower-nacelle with MR TVA system, assuming PMR force as the control input. Using this algorithm, the optimal control signal PMR, which is the force generated by an MR damper, will be obtained. TheMR damper is controlled using a power interface of an analogue type. To induce theMR damper to generate the desired optimal control force, the inverse model of the MR damper is used. This model determines the relationship between the optimal value of the force PMR, actual piston displacement (designated by wx12), actual piston velocity (vx12) and MR damper current, such as: iMR = f(PMR,wx12,vx12). However, the displacement signals of the tower-nacelle with MR TVAsystemaremeasuredonly – thevelocity signals are not available for theLQRcontroller and MR damper inverse model. To solve this problem, the velocities are replaced by their estimates produced by the Kalman filter, described in detail in the next Section. 4.1. Kalman filter TheKalman filter is used to restore unmeasurable variables vx1 and vx12, required to imple- ment the LQR controller (and also the MR damper inverse model). This method provides for state (assumed in Section 4.3) of tower-nacelle with MR TVA system estimation, considering the measurement and process noises. It should be noted that in the LQG design, two Kalman filters of the same structure are used separately for vx1 and vx12. Consider the following system z(k+1)=Akz(k)+Bku(k)+ϑ(k) y(k) =Hkz(k)+ψ(k) (4.3) where: z= [wx,vx,ax] T is the state vector that includes displacement wx (wx1, or wx12), velo- city vx (vx1, or vx12) and acceleration ax (ax1 or ax12, respectively; acceleration is estimated for future applications), while ϑ(k) and ψ(k) are respectively the process and measurement white noises. As onlywx displacement is beingmeasured, therefore the followingmatricesAk,Bk, andHk of equations (4.3) are considered (Singhal et al., 2012) Ak =    1 T0 T 2 0 0 1 T0 0 0 1    Bk =    0 0 0    Hk = [ 1 0 0 ] Implementation of the LQG controller for a wind turbine... 1115 where T0 is the sampling period of the LQG control algorithm (in the simulations and expe- riments, T0 = 0.001s is assumed). For the calculation purposes, the following values of the covariance matrices Qk,Rk are assumed (r is a constant value, Singhal et al., 2012) Qk =    T50/20 T 4 0/8 T 3 0/6 T40/8 T 3 0/3 T 2 0/2 T30/6 T 2 0/2 T0    Rk = [r] The considered Kalman filter algorithm consists of two basic steps: prediction and correction. The prediction step: ẑ − k =Akẑk – predicted value of the state z, P − k =AkPkA T k +Qk – predicted value of the covariance. The correction step: Kk =P − k HT k (HkP − k HT k +Rk) −1 – gain of the Kalman filter, ẑk = ẑ − k +Kk[wx(k)−Hkẑ − k ] – optimal, estimated value of the state z (wx(k) is themeasured displacement value at kT0 time step), Pk =(I−KkHk)P − k – optimal, estimated value of the covariance (I is the identity matrix). The above algorithm has been implemented in form of a Simulink diagram. The Kalman filter was tested experimentally on the laboratory test rig. The tower-nacelle with MR TVA system was excited with a chirp-type force of amplitude 130N applied at the point x0 (A(F)= 130N). The frequency was changing from 35Hz to 1Hz. Figures 4a and 4b present comparison of the displacements and velocities time responses of the tower-nacelle with MR TVA system determined from the experiment and estimated by theKalman filter. The estimated velocity vx1 is compared to the one calculated by the Euler method. Regarding the displacements, time responses practically coincide (Fig. 4a). Analysis of the velocity signals (Fig. 4b) shows the advantage of the Kalman filter over the simple-differential velocity calculation method. Fig. 4. Comparison of the time responseswx1 (a) and vx1 (b) 4.2. Linearization of the wind turbine tower-nacelle model The purpose of the linearization procedure is to obtain a discrete linear model of tower- nacelle system only. The linearization procedure has been carried out using Ident tool from MATLABOptimization toolbox. Themain parameters for a black-box linearmodel of the Ident tool are set as follows (Ljung, 2015): • Model structure: general linear state-space model of the 4th order. 1116 M. Rosół, P. Martynowicz • Focus: simulation which approximates dynamics of the model (the transfer function from measured inputs to outputs) with a norm that is given by the input spectrum. • Estimating method: prediction-error minimization (PEM). To start the linearization procedure, the reference data are required. These data, describing the relationship between wx1, vx1, wx0, vx0 and the MR damper force PMR, may be obtained by performing a simulation test using the nonlinear model of the tower-nacelle system with PMR force as the input. Such a test has been carried out for free vibration of the tower-nacelle model with non-zero initial conditions for the wx1, vx1, wx0 and vx0 signals, and for the PMR force changing randomly every 0.200s in the range of ±10N. An exemplary PMR pattern used for the linearization procedure is shown in Fig. 5. The continuous system was discretised using a zero-order hold with T0 =0.001s. Fig. 5. An exemplary PMR time pattern for the linearization Figures 6 and 7 show comparison of the reference data (nonlinear model responses, solid lines) and linear model responses of the tower midpoint and tower tip (nacelle) displacements and velocities. The numbers indicating the best fit values are given below each figure (the maximum value is 100). As can be observed, the fit level is high – it exceeds the value of 90 for each state variable. Therefore, it can be concluded that the obtained linear model can be used to implement the LQG regulator. Fig. 6. Comparison of the responseswx1 (a) andwx0 (b) of nonlinear and linear models; best fits: 94.52 (a) and 95.73 (b) Implementation of the LQG controller for a wind turbine... 1117 Fig. 7. Comparison of the responses vx1 (a) and vx0 of nonlinear and linear models; best fits: 94.16 (a) and 94.94 (b) 4.3. DLQR controller The Discrete Linear-Quadratic Regulator (DLQR) is a state-feedback controller defined for a discrete-time state-space system. DLQR parameters are calculated by solving the optimal problemcalled thediscreteLQRproblem.This problem is defined for systemdynamicsdescribed by a set of linear differential equations and a quadratic cost function. In this paper, synthesis of a DLQR for a dynamical system described by equations (4.1) is presented. The DLQR optimization problem is solved using dlqr.m function (or dlqry.m) fromMATLAB/SimulinkOptimization toolbox. The dlqr.m function calculates the optimal gain matrix Kd such that the state-feedback control u(k) (optimal MR damper force PMR) u(k) =−Kdx(k) (4.4) with the assumed state x(k) = [wx1(k),vx1(k),wx12(k),vx12(k)] T (wx0 and vx0 are omitted here being proportional towx1(k) and vx1(k), respectively, for the 1st bendingmode frequency neighbourhood regarded), minimizes the quadratic cost function J = ∞∑ k=1 [xT(k)Qx(k)+uT(k)Ru(k)+2xT(k)Nu(k)] (4.5) where:Q=QT ­0,R=RT >0 andN=NT ­0. The DLQR parameters are calculated for the following forms of the matrices Q and R, occurring in quality factor Q= diag[100,10,100,10] (4.6) with differentR values as shown below (denoted by the LQG1), and Q= diag[3000,300,300,30] (4.7) with R equal to 0.0005 (denoted by LQG2). The N matrix is set to zero for both LQG1 and LQG2. The element values of matrices (4.6) and (4.7) are tuned with emphasis put on the stabili- zation of the tower-nacelle system position and limitedMRdamper stroke. Hence,Q11 andQ33 elements of matrices (4.6), (4.7) are ten times greater than elements Q22 andQ44, respectively, while tower-nacelle stabilization purposedominates overMRdamper stroke limitation forLQG2 concept. Themaximum control value is limited by the matrixR. 1118 M. Rosół, P. Martynowicz As the results of calculations, the following values of the DLQR gain Kd are achieved. — LQG1 controller Kd =    [ −74.3077 −11.0044 −65.6904 275.6049 ] for R=0.0001 [ −45.75 −5.32 −46.26 184.99 ] for R=0.0002 [ −23.7730 −2.2100 −28.3578 104.9820 ] for R=0.0005 —LQG2 controller Kd = [ −179.81 −59.13 −29.48 289.19 ] In the next step, the DLQR controller and Kalman filter are integrated, forming the LQG controller. The integration stage has been executed according to the scheme shown in Fig. 8 (carets indicate estimates). Fig. 8. Structure of the integrated LQG controller 5. Simulation analysis Within the scope of the simulation analysis, time and frequency characteristics have been ob- tained. The latter, determined for sine excitation series applied either to the nacelle (excitation force amplitude equal to A(P) = 61N) or to the tower midpoint (A(F) = 150N) are presented below (Figures 9 and 10, see headers). Figures 9a and 10a present output frequency response functions of amplitude A(wx0). Figures 9b and 10b show the output frequency response func- tions of amplitude A(wx1). The information present in the legends of all the following figures has the respectivemeaning: ‘inv’ confirms incorporation of theMRdamper inversemodel, while ‘Fr’ refers to linear guides friction force (of ca. 1N) compensation by the MR damper (both of them as described in Section 2). As can be observed, the A(wx1) amplitude output frequency response functions present ca. three times greater values than the respective A(wx0) functions. For LQG1 case, an increase in the control weighting value R results in lower feedback gain vector Kd modulus, and so in less stiffness and less damping present between the protected structure and the absorber. This, in turn, is apparent as higher two maxima amplitudes and lower in-between the two Implementation of the LQG controller for a wind turbine... 1119 maxima response.TheLQG2case, characterisedwithhigherweights inQ (especially theweights concerning the tower tip displacementwx1 andvelocity vx1) andR=0.0005, produces frequency response functions similar to LQG1 controller with the lowest R (R=0.0001). Fig. 9. (a)A(wx0) and (b) A(wx1) output frequency responses;A(P)= 61N Fig. 10. (a)A(wx0) and (b) A(wx1) output frequency responses;A(F)= 150N 6. Experimental analysis The experimental analysis comprised determination of time and frequency characteristics. The first was a free response test of displacement wx1, obtained for the MR TVA system with se- lected LQG controllers and the system with MR TVA ‘locked’ (Fig. 11). According to the Fig. 11. Time series of displacementwx1 – free response of theMRTVA systemwith LQG controllers and the systemwithMRTVA ‘locked’ constraints of the laboratory facility, the frequency characteristics (Figs. 12-14) were determi- ned for sine excitation series applied either to the nacelle (A(P) = 61N; LQG1 system with R = 0.0001 and R = 0.0002) or to the tower midpoint (A(F) = 150N; LQG1 and LQG2 sys- tems with R = 0.0005). Figures 12a and 13a present the output frequency response functions 1120 M. Rosół, P. Martynowicz of A(wx0) amplitude. Figures 12b and 13b present the output frequency response functions of A(wx1) amplitude.Figure 14presents output frequency response functions ofA(wx0) andA(wx1) amplitudes for the systemwithout TVA (MR TVA ‘locked’). Fig. 12. (a)A(wx0) and (b) A(wx1) output frequency responses;A(P)=61N Fig. 13. (a)A(wx0) and (b) A(wx1) output frequency responses;A(F)= 150N Fig. 14.A(wx0) andA(wx1) output frequency responses;A(F)= 150N Observing the free response plots, one can conclude that LQG1 algorithm with R=0.0001 andLQG2 algorithmwithR=0.0005 provide the best vibration attenuation properties. Regar- ding the frequency analysis, as for simulations, theA(wx1) amplitude output frequency response functions exhibit ca. three times greater values than the respective A(wx0) functions. LQG1 algorithmwithR=0.0001 andR=0.0002 produces comparable results for the excitation series applied to thenacelle, with twomaximaapparent forR=0.0002 case. For the excitation applied to the tower midpoint, LQG1 and LQG2 algorithms (both with R = 0.0005 weighting value) collation may suggest the latter solution (LQG2) to be preferable, with noticeably lower two maxima amplitudes; however, in-between the twomaxima range is better attenuated for LQG1. Implementation of the LQG controller for a wind turbine... 1121 7. Conclusion Theobtained results prove the effectiveness of theLQGcontroller for the considered application. On the basis of the laboratory experiments, almost 10-fold reduction of the displacement wx1 amplitude has been observed for the LQG2 system in comparison to the system without MR TVA(TVA ‘locked’) – see Figs. 13b and 14. Implementation of the LQGcontroller combines the benefits of the LQR state feedback (including system linearization) with noise compensation by the Kalman filter. Based on the analyses, simulations and laboratory model measurements presented in this paper, and considering force scale factor determination (Snamina and Martynowicz, 2014) in combination with the previous results (Snamina et al., 2014), direct calculation of demanded control signal for a real-world full scale vibration reduction system/MRTVAispossible aswell as the calculation of the real-world wind turbine structural deflection and acceleration amplitudes, as for Vensys 82 plant regarded (Martynowicz, 2015). Acknowledgment This work has been supported by the Polish National Science Centre (Narodowe Centrum Nauki) under project No. 2286/B/T02/2011/40. References 1. Asai T., Spencer Jr. B.F., 2014,Model-free algorithms for seismic response control employing controllable dampers,Proceedings of the VI World Conference on Structural Control and Monito- ring, Barcelona 2. Asai T., Spencer Jr. B.F., Iemura H., Chang C.-M., 2013, Nature of seismic control force in acceleration feedback, Structural Control and Health Monitoring, 20, 789-803 3. AthansM., 1971, The role and use of the stochastic linear-quadratic-Gaussianproblem in control system design, IEEE Transaction on Automatic Control, 16, 6, 529-552 4. Bak C., Bitsche R., Yde A., KimT., HansenM.H., Zahle F., GaunaaM., Blasques J., Dossing M., Wedel-Heinen J-J., Behrens T., 2012, Light rotor: the 10-MW reference wind turbine,European Wind Energy Association Annual Event, 16-19.04, Copenhagen, Denmark 5. Butt U.A., Ishihara T., 2012, Seismic load evaluation of wind turbine support structures con- sidering low structural damping and soil structure interaction,European Wind Energy Association Annual Event, 16-19.04, Copenhagen, Denmark 6. COMSOL AB, 2008, COMSOL Multiphysics MATLAB Interface Guide, COMSOL Version 3.5a, November 7. Den Hartog J.P., 1985,Mechanical Vibrations, Dover Publications, Mineola, NY 8. Dyke S.J., Spencer Jr. B.F., SainM.K.,Carlson J.D., 1996a,Anew semi-active control de- vice for seismic response reduction,Proceedings of the 11th ASCEEngineeringMechanics Specialty Conference, 886-889 9. Dyke S.J., Spencer Jr. B.F., Sain M.K., Carlson J.D., 1996b, Modeling and control of magnetorheological dampers for seismic response reduction, Smart Materials and Structures, 5, 5, 565-575 10. Enevoldsen I., Mork K.J., 1996, Effects of vibration mass damper in a wind turbine tower, Mechanics of Structures and Machines, 24, 2, 155-187 11. Favoreel W., De Moor B., Gevers M., Van Overschee P., 1999, Closed loop model- -free subspace-based LQG-design, Proceedings of the 7th Mediterranean Conference on Control and Automation (MED99), Haifa, Israel, 1926-1939 1122 M. Rosół, P. Martynowicz 12. Hansen M.H., Fuglsang P., Thomsen K., Knudsen T., 2012, Two methods for estimating aeroelastic damping of operational wind turbine modes from experiments,European Wind Energy Association Annual Event, 16-19.04, Copenhagen, Denmark 13. Hjalmarsson H., Gevers M., Gunnarsson S., Lequin O., 1998, Iterative feedback tuning: theory and applications, IEEE Control Systems Magazine, 18, 26-4. 14. Jain P., 2011,Wind Energy Engineering, McGraw-Hill 15. Jelavić M., Perić N., Petrović I., 2007, Damping of wind turbine tower oscillations through rotor speed control,EVER Conference 2007, March 29-April 1, Monaco 16. Kawamura Y., 1998, Direct construction of LQ regulator based on orthogonalization of signals: Dynamical output feedback, Systems and Control Letters, 34, 1-9 17. Kciuk S., Martynowicz P., 2011, Special application magnetorheological valve numerical and experimental analysis, [In:]Diffusion and Defect Data – Solid State Data. Pt. B, Solid State Phe- nomena, Vol. 177: Control Engineering in Materials Processing, 102-115 18. Kirkegaard P.H. et al., 2002, Semiactive vibration control of a wind turbine tower using an MR damper, Structural Dynamics EURODYN 2001, H. Grundmann (Edit.), CRCPress 19. Koo J.H., Ahmadian M., 2007, Qualitative analysis of magneto-rheological tuned vibration ab- sorbers: experimental approach, Journal of Intelligent Material Systems and Structures, 18 20. Łatas W., Martynowicz P., 2012,Modelling of vibration of the tower-nacelle system of a wind power plant with a dynamic damper (in Polish),Modelowanie Inżynierskie, 13, 44, 187-198 21. Ljung L., 2015, System Identification ToolboxTM. User’s Guide, MATLAB&Simulink R2015a, TheMathWorks Inc., USA 22. Lord Rheonetic, 2002,MRControllable Friction Damper RD-1097-01 Product Bulletin 23. MartynowiczP., 2014a,Developmentof laboratorymodel ofwind turbine’s tower-nacelle system with magnetorheological tuned vibration absorber, Solid State Phenomena, 208, 40-51 24. Martynowicz P., 2014b, Wind turbine’s tower-nacelle model with magnetorheological tuned vibrationabsorber–numerical andexperimental analysis,6WCSCM:Proceedings of the 6thEdition of the World Conference of the International Association for Structural Control and Monitoirng (IACSM), 15-17.07, Barcelona, Spain 25. Martynowicz P., 2015, Vibration control of wind turbine tower-nacelle model with magnetorheological tuned vibration absorber, Journal of Vibration and Control, doi: 10.1177/1077546315591445 26. MartynowiczP., 2016,Studyof vibration controlusing laboratory test rig ofwind turbine tower- -nacelle systemwithMR damper based tuned vibration absorber,Bulletin of the Polish Academy of Sciences Technical Sciences, 64, 2, 347359 27. MartynowiczP., SzydłoZ., 2013,Wind turbine’s tower-nacellemodelwithmagnetorheological tuned vibration absorber: the laboratory test rig,Proceedings of the 14th International Carpathian Control Conference (ICCC), 26-29.05, Rytro, Poland 28. Matachowski F., Martynowicz P., 2012, Analysis of dynamics of a wind power plant by making use of Comsol Multiphysics environment (in Polish), Modelowanie Inżynierskie, 13, 44, 209-216 29. Namik H., and Stol K., 2011, Performance analysis of individual blade pitch control of offshore wind turbines on two floating platforms,Mechatronics, 21, 691-703 30. Oh S., Ishihara T., 2013, A study on structure parameters of an offshore wind turbine by excitation test using activemass damper,EWEA Offshore, 19-21.11, Frankfurt 31. Rotea M.A., Lackner M.A., Saheba R., 2010, Active structural control of offshore wind turbines,48thAIAAAerospace SciencesMeeting Including theNewHorizonsForumandAerospace Exposition, 4-7.01, Orlando, Florida Implementation of the LQG controller for a wind turbine... 1123 32. Sapiński B., Martynowicz P., 2005, Vibration control in a pitch-plane suspension model with MR shock absorbers, Journal of Theoretical and Applied Mechanics, 43, 3. 33. SapińskiB.,RosółM., 2007,MRdamper performance for shock isolation,Journal of Theoretical and Applied Mechanics, 45, 1, 133-146 34. Sapiński B., Rosół M., 2008, Autonomous control system for a 3 DOF pitch-plane suspension model withMR shock absorbers,Computers and Structures, 86, 3/5, 379-385 35. Shan W., Shan M., 2012, Gain scheduling pitch control design for active tower damping and 3p harmonic reduction,European Wind Energy Association Annual Event, 16-19.04, Copenhagen, Denmark 36. Singhal T., Harit Akshat, Vishwakarma D.N., 2012, Kalman filter implementation on an accelerometer sensor data for three state estimation of a dynamic system, International Journal of Research in Engineering and Technology, 1, 6, ISSN 2277-4378 37. Snamina J.,Martynowicz P., 2014,Prediction of characteristics of wind turbine’s tower-nacelle systemfrominvestigationof its scaledmodel,6WCSCM:Proceedings of the 6thEdition of theWorld Conference of the International Association for Structural Control and Monitoirng (IACSM), 15-17.07, Barcelona, Spain 38. Snamina J., Martynowicz P., Łatas W., 2014, Dynamic similarity of wind turbine’s Tower- -nacelle system and its scaledmodel, Solid State Phenomena, 208, 29-39 39. Snamina J., Sapiński B., 2011, Energy balance in self-powered MR damper-based vibration reduction system,Bulletin of the Polish Academy of Sciences Technical Sciences, 59, 1, 75-80 40. TMS, 2010, 60 LbfModal Shaker, TheModal Shop Inc. 41. Tsouroukdissian A., Carcangiu C.E., Pineda Amo I., Martin M., Fischer T., Kuhnle B., Scheu M., 2011, Wind turbine tower load reduction using passive and semiactive dampers, European Wind Energy Association Annual Event, Brussels 42. Wang Y., Dyke S., 2013, Modal-based LQG for smart base isolation system design in seismic response control, Structural Control and Health Monitoring, 20, 5, 753-768 Manuscript received May 13, 2015; accepted for print January 27, 2016