Professional Documents
Culture Documents
Vol. , No. ,
Bilal Mokrani∗
University of Liverpool, Liverpool, England L69 GH, United Kingdom
Francesco Palazzo†
Polytechnic University of Turin, 10129 Turin, Italy
and
John E. Mottershead‡ and Sebastiano Fichera§
University of Liverpool, Liverpool, England L69 GH, United Kingdom
DOI: 10.2514/1.J057855
This paper presents the first experimental study of multiple-input multiple-output (MIMO) active vibration
suppression by pole placement using the receptance method. The research is based on a purpose-built modular
flexible wing equipped with leading- and trailing-edge control surfaces and two displacement sensors for measuring
its position. The MIMO controller has the advantage of being designed entirely on frequency response functions,
which are measured between the control surface position (control inputs) and the structural displacements (outputs),
and include the actuator dynamics. There is no requirement to evaluate or to know the structural M, C, and K
matrices or the aerodynamic loads; and the formulation eliminates the need for a state observer. The controller is first
implemented numerically and then experimentally on the aeroelastic system. Both frequencies and damping are
assigned (together or independently) for the first two vibration modes. The research includes a procedure for
assessing the control effort required and demonstrates an effective means of increasing the flutter margin while the
effort is minimized.
wing model with four control surfaces (two leading-edge and two Table 1 MODFLEX main
trailing-edge surfaces) and four accelerometers (two at the wing tip specifications
and two at midspan). A reduced-order model for state-space control Wing data Dimension
implementation was tuned using wind-tunnel results, and the
Wing span 1m
resulting controller was found capable of damping the wing-bending Chord c 0.3 m
and torsion modes while remaining stable within the flight envelope. Airfoil NACA0018
Bendikson [12] introduced an energy-based approach whereby the Mass axis position 0.5 × c
system is forced into a stable region of the phase space by altering the Flexural axis position 0.5 × c
aeroelastic mode (bending–torsion amplitude ratio and phase) using
minimal control effort. More recently, Zhang et al. [13] considered
the effect of actuator dynamics on the critical flutter velocity, as did
Singh et al. [14] using the method of receptances [2,24]. Mx Cx_ Kx Bu f a d (1)
The structural-modification theory, from which the receptance
method is derived, has been available for many decades [15–17] but where dn×1 represents an external force vector; um×1 is a vector of m
only quite recently became practically applicable to physical control input forces, and B b1 b2 : : : bm is a n × m topology
engineering structures [18,19]. The modern development of the matrix, describing the control force distribution over the structure.
“transfer path analysis” [20], enabling the effects of assembly (or For aeroelastic systems, f a represents the vector of generalized
disassembly) of components to be computed using vibration aerodynamic forces that can be computed, in the reduced-frequency
measurements, is based on similar principles. Active modification, domain k, by using any numerical method for solving the
using the method of receptances, was first proposed by Ram and potential flow.
Mottershead [2] for structures equipped with a single control input and
several outputs. The method, based on proportional and derivative
A. Experimental Rig
output feedback, relies only on the measured receptances to determine
the controller gains that guarantee the placement of the system poles at The MODFLEX wing is made of four sectors, with a NACA0018
the desired locations. The number of complex pole pairs that can be profile, mounted on a flexible beam and described in Table 1 and
assigned is equal to the number of available sensors. The receptance Fig. 1. The main spar is made of aluminum alloy, and it is the only
method has significant advantages over conventional matrix-based structural element of the test rig. By accurately shaping the cross
methods, such as those described in the first paragraph. In particular, section of the main spar, the desired flexural, torsional, and in-plane
there is no need to know or to evaluate the structural matrices (M, C, K) stiffnesses have been achieved (in order to reach a flutter speed well
or the aerodynamic coefficient matrices. In addition, the use of an within the 0–20 m∕s range of the wind-tunnel facility of the
observer to estimate unmeasured states is unnecessary because the University of Liverpool). The main spar is covered by four sectors
formulation is in terms of receptance rather than dynamic stiffnesses. that provide the correct aerodynamic shape. Each sector is connected
Over the most recent decade, the method has been increasingly applied to the main spar by two pins at the midspan of the sector so as not to
within the research community, leading to several demonstrations alter the stiffness distribution of the overall model. The center of
ranging from aeroservoelastic systems [21], flexible structures [22], and gravity and the flexural axis were designed to be coincident on the
vibroacoustic control [23]. Recently, the receptance method was main spar. The wing model is fully constrained at the built-in end.
generalized for multiple-input multiple-output (MIMO) systems by The fourth sector is equipped with leading- and trailing-edge
Ram and Mottershead [24], and its efficiency was demonstrated aerodynamic control surfaces. Each aerodynamic control surface is
numerically in the active aeroelastic control of a delta wing with multiple actuated with two identical MAXON 60 W brushless motors, each
trailing-edge control surfaces [25]. It was also extended to encompass with a planetary gearbox and an angular position encoder. Two
nonlinear vibration control [26,27]; but, until now, the MIMO contactless laser sensors (Keyence LK-500; mounted on the top of the
receptance method has not been demonstrated experimentally. The wind-tunnel test section) measure the absolute position of the wing at
reader who requires further information on research into active flutter two distinct points on the third sector, as depicted in Fig. 2.
suppression in aircraft is referred to Ref. [28] for an extensive review.
In this paper, the first experimental validation of the MIMO B. Finite Element Model
formulation is demonstrated using a modular flexible aeroelastic A finite element model coupled with unsteady aerodynamics was
wing (referred to as MODFLEX) [29], featuring leading- and produced to assist in the design of the MODFLEX aeroelastic wing.
trailing-edge control surfaces and equipped with two position Of course, the model has no direct role in the design of the controller,
sensors. In addition to the advantages of the receptance method which is based entirely on measured receptances (or transfer
described previously, the implementation herein includes the actuator functions). The model consists of a beam-element model with
dynamics in the transfer-function matrix, thereby eliminating the distributed lumped masses to account for the mass distribution of the
need to model it separately, as in Ref. [14]. The paper is organized sectors and the servomotors used for driving the control surfaces
into five main sections: Section II describes the aeroelastic
MODFLEX rig used for demonstration as well as its numerical
model. Section III recalls the formulation of the multi-input multi-
output receptance method and describes its implementation in
aeroelasticity. Sections IVand V describe, respectively, the numerical
and experimental implementations of the receptance method. Finally,
the outcomes are summarized in the Conclusions section (Sec. VI).
control system will be targeted on the first bending (1B) and the first
torsional (1T) modes.
D. Dynamic Model
A state-space model of the MODFLEX is developed for
preliminary numerical validation of the receptance-based controller.
The structural and aerodynamics matrices are extracted and
rearranged in the style of Eq. (1), where f a qHam k; Mx and
Ham is the aerodynamic influence coefficient matrix. The state-space
model is arranged in such a way that the inputs are the torques applied
at the hinges of the control surfaces and the outputs are the angular
Fig. 2 MODFLEX experimental setup. positions of the control surfaces and the absolute vertical position of
the two degrees of freedom located at the third sector.
The model is reduced to include only the states associated with the
first bending, torsional, and in-plane flexible modes (1B, 1T, and 1P
(developed in MD.Nastran). Aerodynamic panels are added to the in Table 2), four states associated with the two rigid-body modes of
finite element model to solve the potential flow via the doublet lattice the control surfaces, and five aerodynamic states. The aerodynamic
method. Structural modal damping of 2% was assumed. states are obtained by fitting the matrix of aerodynamic influence
A preliminary modal analysis campaign was conducted for coefficients Ham using Roger’s classical expansion [3]:
checking the behavior of the experimental model against the
X
N
numerical one. In Table 2, the numerical data (in terms of frequency p
ωi and damping ξi ) are compared with the experimental data for an f a q D0 D1 p D2 p2 Ei x (2)
p − βi
airspeed of V 10 m∕s; only the first two modes are relevant for the i1
Fig. 3 Numerical vs experimental results. Velocity vs frequency and velocity vs damping (−2ξ) diagrams of the MODFLEX wing.
Fig. 4 Block diagram of the HAC–LAC control architecture: a) detailed block diagram; and b) equivalent block diagram, with f d Rsd1 ;d2 T . The
receptance-based controller, presented in the following section, is Cs FT s GT .
E. Low Authority Control of the Aerodynamic Surfaces transfer function, but it is in fact applicable to transfer functions of
The controller of the aeroelastic wing, illustrated in Fig. 4, is built any form. In the present case, the theory is presented in terms of the
in two cascaded control loops: 1) a low authority control (LAC) loop matrix of transfer functions Rs relating the encoder control surface
that controls the angular position of the aerodynamic control positions uLE ; uTE T as inputs to the output laser displacements
surfaces, and 2) a high authority control (HAC) loop in which any measurements x1 ; x2 T at two locations on the wing, as shown in
active aeroelastic controller can be implemented; in this case, it is the Fig. 2. The receptance method proposes a systematic control design
receptance-based controller for partial pole placement as described in procedure for pole placement, without any use of state observers or
Sec. III. The purpose of the LAC loop is to dynamically control the underlying numerical models. The number of pole pairs that can be
position of the control surfaces in order to generate the desired torque placed is equal to the number of sensors. Thus, for the control of the
at the hinges of the control surfaces. In addition, it stabilizes the aeroelastic wing, with two sensors, we aim to modify the eigenpairs
leading-edge flap, which is inherently unstable due to the rigid-body associated with the two first modes.
dynamics of the leading-edge control surface facing the wind flow. It Consider the system of Eq. (4), where the matrices M, ~ C,~ and K~ are
should be noted that d d1; d2 T in Fig. 4a represents a disturbance inclusive of the aerodynamic terms. The matrix Hs in Fig. 4 is then
input. Furthermore, the aeroelastic control is more convenient if the given by Ms~ 2 Cs~ K ~ −1. The input–output transfer function in
position of the control surfaces is directly commanded instead of the question is not Hs but Rs, where
torque. Therefore, in order to stabilize the system and use the position
of the aerodynamic control surfaces as a command, a low authority yx Rsu (5)
control loop is implemented for each control surface. A proportional–
integral–derivative (PID) controller uses the angular position of the In the present example, input matrix B is already incorporated in
control surface (measured with a position encoder) to calculate the Rs such that every output is coupled to both inputs. Assuming a
appropriate torque to apply in order to maintain the control surface at proportional and derivative feedback control, i.e.,
a desired position (internal loop in Fig. 4a). The PID controllers were
tuned at zero airspeed and have a bandwidth of 30 Hz for the trailing u FT x_ GT x (6)
edge and more than 100 Hz for the leading edge; both were
experimentally checked up to 15 Hz. Their dynamics are well above The receptance method offers a straightforward procedure aimed
the aeroelastic modes of interest. With such a configuration, the input at determining the feedback gains F and G by using only the
of the system becomes the control surface angular position measured transfer function Rs; it can be applied by following the
u uLE ; uTE T , and the outputs remain the absolute positions five steps in the flowchart of Fig. 5, as will be described in the
measured at two different points on the flexible wing: yx x1 ; x2 T . following section.
A. Method
III. Pole Placement with the Receptance Method The theory presented here is for the general case of a system with
In this section, the multi-input multi-output receptance method is 2n poles; although, specifically, n 1 or 2 in the present example.
recalled. The original theory [24] was given in terms of the receptance Therefore, let μk , k 1; 2; : : : ; p, be the set of desired poles to be
MOKRANI ET AL. 5
Fig. 6 Numerical model: open-loop frequency response functions Rs at wind speed of V 10 m∕s: a) x1 ∕uLE ; b) x1 ∕uTE ; c) x2 ∕uLE ; and d) x2 ∕uTE .
avoid spillover. As discussed previously, the parameters αμk ;j are f d Rsd1 ; d2 T to the control commands u uTE ; uTE T .
selected to minimize the control effort by numerically exploring all From Fig. 4, the following can be shown:
the possibilities.
u CsI − RsCs−1 f d Gd sf d (13)
Fig. 7 Measured vs fitted open-loop FRFs Rs at wind speed of V 10 m∕s. Inputs are angular positions of the control surfaces (uLE , uTE ). Outputs are
position of the flexible wing (x1 , x2 ).
Fig. 8 FRF of x1 ∕d1 , where mode 1B is targeted: a) open loop (dotted line), ξ1 × 2 (red dashed line), and ξ1 × 3 (blue solid line); and b) open-loop (dotted
line), ω1 × 0.8 (red dashed line), and ω1 × 1.2 (blue solid line).
Fig. 9 FRF of x1 ∕d1 , with open loop (dotted line) and closed loop (red solid line): a) ξ1 × 2 and ξ2 × 2, and b) ω1 × 0.8 and ω1 × 1.2.
8 MOKRANI ET AL.
Fig. 10 Maximum singular value σ of Gd s with various choice of the parameters αk . The control is targeted for increasing a) ξ1 × 2 and b) ω1 × 1.2.
Table 3 Closed-loop control test campaign conducted on the flexible wing: new pole location
vs desired pole location
Desired pole location New pole location
Test μ1;2 μ3;4 μ1;2 μ3;4
1 ξ1T × 2 −1.25 17.8j −1.76 25.8j −1.42 17.8j −1.78 25.3j
2 ξ1T × 3 −1.25 17.8j −2.63 25.7j −1.41 17.7j −2.7 25.5j
3 ξ1B × 2 and ξ1T × 2 −2.5 17.7j −1.76 25.8j −2.33 17.7j −1.72 25.1j
4 ω1T × 1.2 −1.25 17.8j −1.05 31j −1.67 17.8 −1.8 31.3j
5 ω1B × 0.9 and ω1T × 1.1 −1.13 16j −0.97 28.4j −1.26 16j −1.36 28.2j
modes, whereas tests 4 and 5 are targeted for modifying their ξ1T 5.7% instead of 3.4%); this may be associated to the fact that
resonance frequencies; the associated FRFs (x1 ∕d1 ) with and without the control surface deflection is very high, which possibly results in
control are shown in Figs. 11 and 12, respectively. The associated additional aerodynamic damping.
feedback gains are showed in Table 4.
Clearly, in all of the tests, the imaginary part is an order of
magnitude greater than the real part. For the first three tests, the error A. Control Effort
on the pole location is negligible, and it is mainly associated with the The control effort of the system has been measured by
smaller real part of the pole (i.e., the damping component of the pole). calculating the singular values of the transfer-function matrix
Similarly, for tests 4 and 5, one can consider the error on the new pole relating the input vector f d Rsd1 ; d2 T and the computed
locations to be negligible for the imaginary part of the pole (and the commands uLE ; uTE T , as mentioned earlier and described in
resonance frequency). However, especially for these two tests, the Appendix B. Figure 13a shows the maximum singular values σ
error in the damping is relatively high and may reach 66% (in test 4, associated with tests 1, 2, and 3, where the control is targeted only
Fig. 11 Measured FRF x1 ∕d1 with and without control (dotted line): a) tests 1 (red dashed line) and 2 (blue solid line); and b) test 3 (red solid line).
Fig. 12 Measured FRF x1 ∕d1 with (red solid line) and without control (dotted line): a) test 4; and b) test 5.
MOKRANI ET AL. 9
Table 4 Computed feedback gains Table 5 Flutter speed with various control
configurations
Test F G
Test Flutter speed, m∕s
0.4 0.6 −7.5 11.4
1
−0.6 −0.4 5.2 −16.7 0 Open-loop 13.5
1 ξ1T × 2 15.5
0.8 0.2 −11.9 21.9
2 2 ξ1T × 3 16.5
−1.2 0.3 6.7 −24.9
3 ξ1B × 2 and ξ1T × 2 15.5
−1.2 3.7 5.9 −22.8 4 ω1T × 1.2 14
3
−1.4 1.2 11.9 −33.8 5 ω1B × 0.9 and ω1T × 1.1 15
1.7 −3 65.2 −115.2
4
−0.3 0.6 −83.2 147.3
−2.3 0.4 52.8 −130.7 corresponds to test 2 where the damping of the torsional mode is
5
−1.8 0.1 −28.3 16.8
tripled.
The experimental velocity vs frequency/damping (where the
damping is expressed as −2ξ) diagrams are plotted in Fig. 15, with
for the damping. When only the second-mode (1T) is targeted, the and without control. The control is tuned to triple the damping of the
control effort reaches a maximum around this mode, whereas there first torsional mode (test 2). As imposed by the control design, only
is no peak around the first mode (1B); this is due to the nature of the damping of the first torsional mode is increased, whereas the
the receptance-based control, which works in a similar way to a resonance frequencies of the 1B and 1T modes remain almost
modal filter, as expressed by the constraint Qk f Tk gTk T 0 identical to the open-loop system. It is clearly seen that the control
in Eq. (7). shifts down the damping curve of the torsional mode, which is
Figure 13b shows σ for tests 4 and 5, where the control is targeted responsible for flutter. Notice also that, even at rest, without any flow
for shifting the resonance frequencies of the modes of interest: 1B and around the wing (i.e., V 0 m∕s), the control still provides
1T. The control effort is significantly higher as compared to the tests significant damping to the targeted mode; this is possible due to the
when only the damping is increased, and it was at the limit of the relatively high inertia of the leading-edge control surface, which
control surfaces stroke (with relatively small disturbances). This fact behaves like an inertial actuator.
could also be observed from the calculated gains that are one order of
magnitude higher in tests 4 and 5 than in tests 1, 2, and 3, cf. Table 4.
Once again, when only a single mode is targeted, the control effort is
concentrated around this mode. Recall that, for the various control
tests conducted on the aeroelastic wing, the parameters αμk have been
chosen in order to minimize the control effort; thus, without a careful
choice of αμk , the controller would not be feasible.
B. Flutter Control
A flutter test was carried out for the five active control tests
depicted in Table 5. The wing is positioned in the wind tunnel, with
the control surfaces positioned at zero angular position (aligned with
the wing), while the control is turned on. The wind speed is then
increased progressively until the flutter speed is reached. Table 5
depicts the flutter speed for each control test; test 0 is the reference
case, in which only the LAC loop is active, and the system exhibits a
flutter speed of 13.5 m∕s.
The experimental campaign showed that increasing the damping
of the torsional mode (test 2) is the most efficient way to increase the
flutter speed because the flutter occurs when the damping of this
specific mode becomes negative. Figure 14 shows the time response
Fig. 14 Measured wing position at wind speed close to 13.5 m∕s. The
of the position sensors x1 , near an air speed of 13.5 m∕s, without control is turned on near the 10th second. The control is tuned to triple the
HAC control and when the control is turned on after 10 s; the control damping of the first torsional mode (test 2).
Fig. 13 Control effort: measured σ of the transfer-function matrix between d1 ; d2 T and the computed commands uLE ; uTE T : a) tests #1, #2, and #3;
and b) tests #4 and #5.
10 MOKRANI ET AL.
Fig. 15 Measured a) frequency ω, and b) damping (−2ξ) as a function of the wind speed: open loop (dashed line) and closed loop (solid line). The control is
tuned to triple the damping of the first torsional mode (test 2).
Finally, one should note that the controller has been tuned based on modes are identical (at the output degrees of freedom) does not, in
the measured model at a wind speed 10 m∕s, and it has not been general, lead to the lowest control effort.
adapted as the wind speed evolves. Although the control proves to be 5) A low-pass filter may be used as a practical solution to avoid the
effective, it would be worthwhile for the feedback gains to be adapted spillover of high-frequency modes when the number of sensors is
as a function of the wind speed. limited. This may lead to increased control effort and could be
difficult to apply in the case of closed modes.
6) An increase in the flutter velocity of around 22% was achieved
for the modular flexible wing that formed the object of the test.
VI. Conclusions
In this paper, a pole-placement controller, designed by using the
Multi-Input Multi-Output receptance method, was implemented both
Appendix A: MODFLEX
numerically and experimentally for the first time on an aeroelastic Structural characterization of the MODFLEX numerical/
system. The method considers proportional and derivative position experimental system in terms of frequency, damping, and mode
feedback and relies only on the measured input–output frequency shape are depicted in Table A1 and Fig. A1.
response functions to calculate the controller gains. The efficiency
and accuracy of the method are demonstrated through a set of control Appendix B: Control Effort–Selection of αμk
scenarios applied on an aeroelastic wing where the receptance-based
To evaluate the control effort, we evaluate the transfer-function
aeroelastic control proved capable of assigning both frequencies and
matrix between the disturbance vector f d Rsd1 ; d2 T and the
damping, both individually and together.
commands u uLE ; uTE T . From Fig. 4b, we have the following:
Following the test campaign and the numerical simulations, the
following main concluding remarks may be made: u CsI − RsCs−1 f d Gd sf d (B1)
1) The study demonstrates the merits of the receptance method in
simplifying the control design for active vibration control of flexible where Cs FT s GT is the receptance-based controller, and
structures in general, without any need for a quantitative numerical Rs is the transfer-function matrix relating the control inputs to the
model or the use of state observers. control outputs; it includes the dynamics of the subsystems, the
2) The method applied in the study eliminates the need to model amplifiers, and the filters as depicted in Fig. 4a.
the actuator dynamics separately; it is included in the measured Equation (B1) describes the reaction effort of the controller
input–output transfer function. (i.e., commands to the actuators) to a given disturbance profile. This
3) A numerical search procedure based on singular value effort can be evaluated in several ways; one simple way is to consider
decomposition of the transfer-function matrix between the the mean value of u for a given disturbance vector spectrum. The
disturbance vector f d Rsd1 ; d2 T and the control commands power spectral density matrix of the commands Φu is related to the
u uTE ; uTE T may be used to determine the minimum control power spectral density matrix of the disturbances Φfd by the
effort. following:
4) The parameters αμk , k 1; 2; : : : ; m define the kth closed-loop
mode shape. Selecting αμk such that the closed-loop and open-loop Φu ω Gd ωΦfd ωG d ω (B2)
[24] Ram, Y. M., and Mottershead, J. E., “Multiple-Input Active Vibration Solids, Jan. 2018.
Control by Partial Pole Placement Using the Method of Receptances,” doi:10.1177/1081286517744601
Mechanical Systems and Signal Processing, Vol. 40, No. 2, 2013, [28] Livne, E., “Aircraft Active Flutter Suppression: State of the Art and
pp. 727–735. Technology Maturation Needs,” Journal of Aircraft, Vol. 55, No. 1,
doi:10.1016/j.ymssp.2013.06.008 2018, pp. 410–452.
[25] Singh, K. V., McDonough, L. A., Kolonay, R., and Cooper, J. E., doi:10.2514/1.C034442
“Receptance-Based Active Aeroelastic Control Using Multiple Control [29] Fichera, S., Jiffri, S., and Mottershead, J. E., “Design and Wind Tunnel
Surfaces,” Journal of Aircraft, Vol. 51, No. 1, 2014, pp. 335–342. Test of a Modular Aeroelastic Flexible Wing (MODFLEX),”
doi:10.2514/1.C032183 International Conference on Noise and Vibration Engineering
[26] Zhen, C., Jiffri, S., Li, D.-C., Xiang, J.-W., and Mottershead, J. E., (ISMA2016), Katholieke Universiteit Leuven, Leuven, Belgium,
“Feedback Linearisation of Nonlinear Vibration Problems: A New Sept. 2016, Paper 179.
Formulation by the Method of Receptances,” Mechanical Systems and [30] Peeters, B., Lowet, G., Van Der Auweraer, H., and Leuridan, J., “A New
Signal Processing, Vol. 98, Jan. 2018, pp. 1056–1068. Procedure for Modal Parameter Estimation,” Sound and Vibration,
doi:10.1016/j.ymssp.2017.05.048 Vol. 38, No. 1, 2004, pp. 24–28.
[27] Lisitano, D., Jiffri, S., Bonisoli, E., and Mottershead, J. E.,
“Experimental Feedback Linearisation of a Non-Smooth Nonlinear C. Pettit
System by the Method of Receptances,” Mathematics and Mechanics of Associate Editor