Professional Documents
Culture Documents
Simon 2015
Simon 2015
a r t i c l e i n f o a b s t r a c t
Article history: This work investigates the stability of a fxLMS controller for active wave cancelation of broad-band Tollmien–
Received 20 April 2015 Schlichting disturbances in a flat plate boundary-layer with a single DBD plasma actuator. In particular the
Revised 1 September 2015
influence of a changing free stream velocity and the resulting off-design operation of the control algorithm is
Accepted 24 September 2015
analyzed up to an unstable behavior. As the main reason for unstable controller operation in the off-design
case the difference between actual and predicted phase angle of the disturbances at the position of the error
Keywords: sensor is identified. A method for an online adjustment of the secondary-path model to different free-stream
Laminar flow control velocities is presented. Finally a wall-bounded method based on the disturbances phase speed is developed
Active wave cancelation that can cope with changes of the physical secondary path not only due to changes of the free-stream velocity
Plasma actuator
but also due to changes of the pressure distribution. This method enables the extension of the stable operation
fxLMS
range of the control system significantly.
Adaptive secondary path
© 2015 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatfluidflow.2015.09.003
S0142-727X(15)00115-0/© 2015 Elsevier Inc. All rights reserved.
B. Simon et al. / International Journal of Heat and Fluid Flow 56 (2015) 344–354 345
Fig. 2. (a) Disturbance source design: tubes are connected to a 0.2 mm wide slit. (b)
Surface hot-wire sensor in a plastic case.
where α is the step size and r(n) the discretized reference sensor sig-
Fig. 3. Spatial distribution of the body-force fx and electric circuit of the plasma actu- nal. If the error sensor is not installed at the actuator position, the re-
ator. sulting phase-angle error due to Hec can lead to an unstable controller
behavior of the LMS. The sensor has to be positioned at least 30 mm
downstream of the actuator in order to avoid electromagnetical inter-
ference and/or damaging the sensor. To account for this displacement
the physical path Hec is modeled by a FIR filter Ĥec , as the lower part
of Fig. 1 shows. The discretized input r(n) is then filtered by Ĥec and
the resulting r (n) serves as an input for the LMS algorithm to adapt
the filter w. In the literature r is often called x and the additional fil-
tering with Ĥec causes the name fxLMS (filtered x LMS) (Elliott, 2000).
Eq. (3) shows the algorithm for the fxLMS controller:
The output of the filter w(n) is the control signal c calculated with
the upstream sensor signal r(n) and therefore is an estimator for the
transmission path Hcr (to c from r). After switching on the controller
Fig. 4. Operation modes of the plasma actuator. the filter w(n) is adapting starting with zero for all filter coefficients.
Depending on the step size α it can take several seconds up to one
minute until the filter is fully adapted for a steady point of operation.
modulated such that velocity disturbances can be canceled out (Kurz For the experiments presented in this paper α is chosen to be α =
et al., 2013). On the time scales of the boundary-layer flow the plasma 10−3 which has proven to be a good compromise between adaptation
actuator force is unidirectional and always directed in the down- speed and controller stability.
stream direction. Therefore a net force remains in average which The fxLMS algorithm is commonly used for flow control (Kotsonis
modifies the stability properties of the boundary layer additionally et al., 2013; Peltzer et al., 2009) because of its robustness and ability
to the intended wave cancelation. This mean body-force can im- to adapt to slight changes of the transmission path. In this manuscript
prove the stability properties of the flow (boundary-layer stabiliza- the algorithm has been expanded by an adaptive secondary path
tion) whereas the modulated PA force is able to attenuate velocity model Ĥec to increase the range of stable controller operation for
fluctuations in the boundary layer (active wave cancelation). One can changes in wind-tunnel speed UWT . Different ways of adapting the
say, that a hybrid operation mode is applied, if the steady mean part secondary path model by a function P and experimental results are
changes the stability properties of the mean flow significantly and the presented in Section 5.
PA is additionally modulated in time for active wave cancelation. The control algorithm is implemented in MATLAB Simulink and
compiled for a dSPACE digital signal processor, consisting of a DS1007
processor board, a high speed A/D board DS2004 as well as a DS2102
2.3. Control algorithm and digital signal processor high-resolution D/A board. The control algorithm and data acquisi-
tion run at 10 kHz, synchronized with the LabVIEW measurement
The flow-control system (Fig. 1) is a SISO (single-input-single- system.
output) system, which consists of a reference sensor r, the error
sensor e and the actuator c. Hxy stands for a transfer function (to actu-
3. Base flow characterization
ator/sensor x, from actuator/sensor y) therefore Her denotes the pri-
mary path (to e, from r) while Hec is the secondary path (to e, from
Hot-wire boundary-layer measurements have been carried out to
c). Thus the secondary path describes the transfer function between
characterize the flow for the reference case at a wind-tunnel speed
the plasma actuator c and the error sensor e while the primary path
of UW T = 12 m/s, which corresponds to a free-stream velocity of the
describes the transmission behavior between reference sensor r and
flat-plate boundary layer of U∞ = 14 m/s due to blockage. The lin-
error sensor e in downstream direction. Experimental investigations
ear stability theory (LST) calculations for the reference case UW T =
showed that there is no significant fluid mechanical upstream effect
12 m/s are shown in Fig. 5, assuming a 2D wave front (spanwise wave
of the plasma actuator on the flow at the reference sensor position.
number β = 0) and a Blasius boundary layer. All stability calculations
The feedback path Hrc (to r, from c) is therefore negligible in the pre-
in this manuscript have been performed with a code provided by
sented flow control system as the disturbances only propagate down-
Reeh (2014).
stream. Therefore the applied control approach can be classified as
Fig. 6 shows the boundary-layer profiles at x p1 = 287 mm, xr =
feed forward type. The fxLMS control algorithm is based on FIR (fi-
335 mm and xe = 408 mm. Three profiles are measured next to the
nite impulse response) filters. The output y(n) of the FIR filter w is a
surface hot-wires p1 , r and e. All profiles have been acquired at 10
linear combination of the last M + 1 input signals x:
kHz sampling rate for 10 s for each data point. Phase averaging allows
M the measurement of a single frequency harmonic TS-wave, which ist
y(n) = wi x[n − i] = wT x(n) = xT (n)w (1) created by the disturbance source d at f = 200Hz. The resulting mea-
i=0 sured waveforms are averaged over 2000 samples for each of the 50
B. Simon et al. / International Journal of Heat and Fluid Flow 56 (2015) 344–354 347
Fig. 5. Neutral stability curve (αi = 0) and N-factor contours for the reference case
UW T = 12 m/s (U∞ = 14 m/s) and zero pressure gradient, spanwise wave number β =0.
Fig. 8. Coherence γ 2 of the primary path Her for different UWT . Fig. 10. Impulse response model for the secondary path Ĥer at different UWT , modeled
by a FIR filter with M = 256 coefficients.
Fig. 9. Upper and lower TS-wave frequency limits, extracted from the coherence γ 2 of
Her .
Fig. 11. Coherence γ 2 of the secondary path Hec for different UWT .
test runs show a rise of the amplitude for all frequencies, which con-
firms the assumption of a change in boundary-layer receptivity. cause the spatial amplification of the disturbance is not taken into ac-
However, the magnitude of the variation of the receptivity process count for αi = 0. However, the spatial amplification is included in the
between different runs is considered to not have an influence on the calculation of the amplification factor N, as done for the plot of Fig. 5.
controller stability. By extracting the bandpass from the contour plot of N for UW T =
12 m/s in Fig. 5 at x = 0.37 m, the bandpass boundaries of 150 Hz < f
3.2. Boundary-layer transmission behavior for variable flow velocities < 300 Hz compare very well to the experiment.
The control success of the fxLMS controller is depended on a good
The transmission behavior of the boundary layer is changing with model Ĥec of the boundary-layers transmission behavior between
the local Reynolds number Rex = U∞ x
ν and pressure gradient. If x is plasma actuator c and error sensor e. All experiments in this doc-
held constant and a Blasius boundary layer is assumed over the whole ument have been conducted in a boundary layer with almost zero
velocity range, the boundary-layers’ transmission behavior is mainly pressure gradient, therefore Ĥec only changes with UWT as the sensor
depending on U∞ or UWT , respectively. Fig. 8 shows the coherence γ 2 and actuator positions are fixed. Fig. 10 shows the impulse response
of the primary path Her , whereas γ 2 close to one implies a linear re- of the secondary path Ĥec , modeled by a finite impulse response (FIR)
lation between the upstream and downstream sensor signals r(t) and filter with M = 256 filter coefficients. It is obtained by operating the
e(t) for a white-noise disturbance excitation. Therefore it shows the high voltage generator with an input voltage c(t), which consists of
relevant frequency band for the active wave cancelation at different a DC offset voltage and an AC PRBS (pseudo random binary signal),
UWT . The dashed parts of the spectra indicate low and high frequency resulting in a modulation of the plasma-actuator body force (Fig. 4).
disturbances, which are not considered to be in the TS-wave band. This broad-band body-force causes random velocity fluctuations in
The control success of the TS-wave attenuation is measured within streamwise direction, which are filtered by the boundary layer while
this frequency band, which makes the definition of these boundaries propagating downstream due to the bandpass characteristic of the
necessary. In Fig. 8 the TS-wave band for UW T = 8 m/s is clearly vis- boundary-layer flow, as the coherence of Ĥec in Fig. 11 demonstrates.
ible whereby it starts at about 100 Hz and ends at 200 Hz. Low fre- In the TS-wave bandpass region a coherence of γ 2 ≈ 0.95 indicates
quency disturbances up to 100 Hz are only slightly damped, which an almost linear behavior of the PA at the given average high-voltage
is verified by a coherence γ 2 ≈ 0.4. The drop for frequencies higher amplitude of V = 9 kV. The impulse-response shape changes as the
than 200 Hz is steeper and ends at a lower level of γ 2 ≈ 0.1. The lower bandpass shifts to higher frequencies, when UWT is increased (Fig. 10).
frequency limit of the TS-wave band is defined at a coherence level of The shape of the impulse-response model mainly relies on the change
γ 2 = 0.7 and the upper limit is defined at a level of γ 2 = 0.1. The of the phase speed c and the amplification of the disturbances in the
solid part of the curves in Fig. 8 lies in between the two limits and boundary layer.
indicates the TS-wave bandpass. Fig. 12 shows the non-dimensionalized (by U∞ ) phase speed c of
The amplified TS-frequency band of the primary path Her , ex- a disturbance at 200 Hz in a Blasius boundary layer at different UWT
tracted with the described method, is shown in Fig. 9 for the whole calculated by LST. The phase speed c changes only slightly in the spa-
velocity range. In addition the lower and upper frequency of the neu- tial region of interest between the first and the last sensor (0.3 m < x
tral stability curve αi = 0 at x = 0.37 m is plotted. The lower and up- < 0.41 m). In order to compare the calculated data from the LST with
per bound of the TS-wave band, extracted from the coherence of the experiment, the calculated wave speed from Fig. 12 is averaged
Her , do not match the neutral stability curve αi = 0 perfectly be- between the sensors p2 and r and plotted as a solid line in Fig. 13.
B. Simon et al. / International Journal of Heat and Fluid Flow 56 (2015) 344–354 349
Fig. 12. Non-dimensionalized phase speed c of the disturbance at f = 200 Hz for dif-
ferent UWT . Fig. 14. Power density spectrum of the error sensor signal e(t) with and without steady
actuator forcing at UW T = 12 m/s.
Fig. 13. Phase speed c of the disturbance at f = 200 Hz, calculated with different sen-
sor pairs for the experiment and corresponding phase speed, calculated with LST at Fig. 15. Secondary path model Ĥec for different high voltage amplitudes V at UW T =
x = 0.37 m. 12 m/s.
The disturbance-wave speed c in the experiment can be determined (on the timescale of the boundary-layer flow). This leads to a slightly
with the phase-angle difference d (in degree) between two sensor fuller boundary-layer profile and a broad-band damping of the dis-
signals with the sensor distance x at a given frequency f: turbances as shown in Fig. 14. The dashed line indicates the power
360◦ density spectrum of the error-sensor signal e(t) with a broad-band
c( f ) = x · f. (5) disturbance caused by the disturbances source while the solid line
d
shows the case for steady forcing at a high voltage V = 9 kV. This off-
The phase-angle difference d is extracted from the measure- set corresponds to the steady force during the experiments presented
ment data with a FFT. For a comparison between LST and experiment, in Sections 4 and 5 and leads to an attenuation of about 2 dB in the
the measured wave speed is shown in Fig. 13 for different upstream TS-wave band. The stability characteristics of the boundary layer are
sensor combinations at f = 200 Hz. The experimentally determined not significantly influenced by this rather small steady force because
curves, calculated with the phase-angle difference between the sen- the attenuation of 2 dB only contributes a small portion to the total
sor pairs, show a similar behavior but the slope differs, leading to control success of about 12 dB with active-wave cancelation (cmp.
a deviation of c ≈ 1 m/s at UW T = 17 m/s. There are two main er- Section 4). The average high-voltage amplitude of V = 9 kV is chosen
ror sources, which may contribute to an error of c. First, the data has to ensure a continuous and stable discharge of the PA during the am-
been acquired at a sampling rate of 10 kHz, which leads to a phase- plitude modulation while a lower amplitude could lead to additional
◦ 200 Hz
angle resolution of 360 ◦
10000 Hz = 7.2 . On the other hand an accuracy turbulence production caused by the discharge onset, which is about
of 1 mm for the positioning of the sensors is assumed. Error bars, V = 5 kV for this actuator setup. Also the PA shows an almost linear
which include both error sources are shown in Fig. 13. The phase relation between the analog input voltage of the high-voltage gen-
speed c calculated by LST differs from the experimental results which erator GBS Minipuls 2.1 and the resulting body-force in the operation
may be owed to the fact that a zero-pressure gradient boundary layer range of 7 kV < V < 11 kV. This linear relation has already been shown
cannot be assumed for the whole velocity range of 7 m/s < UWT < with a coherence γ 2 ≈ 0.95 in Fig. 11 in the TS-wave frequency band.
17 m/s. Nevertheless, the LST calculations give an insight on the fluid- The stabilizing effect of the steady body-force offset causes a
dynamic behavior while the surface hot-wire sensors only can char- change in phase speed c, as shown before with linear stability cal-
acterize the local behavior of the boundary layer at one specific point. culations by Duchmann et al. (2010). The steady force causes a local
acceleration of the wave and therefore a phase-angle difference in
3.3. Transmission behavior for steady plasma actuator force the secondary path Hec . Fig. 15 shows the impulse-response model
Ĥec for different values of V. In contrast to the stretching and scal-
As mentioned before, the plasma actuator has two effects on the ing of the FIR filter coefficients due to a change in wind-tunnel speed
boundary layer. Unsteady force generation can be used to introduce UWT (Fig. 10) the impulse-response curves seem to be shifted in time
velocity fluctuations and steady forcing can change the stability prop- by the changing steady actuator force. In conclusion the steady ac-
erties of the boundary layer. As shown by Kurz et al. (2013) both tuator force can have a significant impact on the secondary path
effects can be applied simultaneously by modulating a steady force Hec in hybrid operation mode. This can not only be caused by dif-
(hybrid mode). However, even without an intentionally added steady ferent voltage amplitudes but also by changing environmental con-
force offset to an unsteady forcing the plasma actuator always adds ditions and a thereby changing discharge behavior and body-force
a net momentum to a flow because its body force is unidirectional production.
350 B. Simon et al. / International Journal of Heat and Fluid Flow 56 (2015) 344–354
Fig. 18. Phase-angle error e caused by different UWT , relative to UW T = 12 m/s. The
solid curves show the experimental data while the dashed curves indicate LST calcula-
Fig. 16. Power density spectra of the error sensor signal e(t) with and without active tions.
wave cancelation at UW T = 13 m/s. The maximum error sensor signal reduction Emax is
indicated with a double-arrow.
tion or even amplification). Because of the unstable behavior of the
controller the measurement has been stopped to prevent actuator
damage and therefore no representative data is available for these
operation points. All colored blocks show measured cases and their
corresponding reduction of the error sensor signal Emax . Red color in-
dicates a stable controller operation and successful TS-wave attenua-
tion whereas yellow blocks indicate a marginally stable controller, for
which the achieved attenuation is almost negligible. The controller
can successfully attenuate TS-waves in the velocity range 11 m/s <
UWT < 15 m/s with a secondary path model, identified at UW T,ident =
12 m/s. The border-line cases UW T = 10 m/s and UW T = 16 m/s show
marginal stable behavior while all other wind-tunnel velocities do
not allow a stable controller operation with this particular model.
Higher UWT allow a wider operation range in terms of wind-tunnel
speed deviation off the design point than lower UWT . This observation
leads to the following more precise discussion of the operating range
of the controller dependent on the phase-angle error, which is caused
by a secondary path model Ĥec,UW T recorded at UWT, ident = UWT .
Fig. 17. Maximum error sensor signal reduction Emax with constant secondary path
4.1. Phase-angle dependent stability of the control algorithm
model (identified at UWT, ident ) and controller performance for different flow speeds
UWT .
The results in Fig. 17 show an unstable behavior of the controller,
when the deviation of UWT from the design case exceeds a certain
4. Stability of the fxLMS controller limit of 2 − 3 m/s. A reason for the unstable controller is the change of
the transmission behavior (stability features) of the boundary-layer
Successful TS-wave cancelation experiments with broad-band flow (Section 3.2), which results in a change of phase speed and am-
disturbances have been documented in the literature before (e.g. plification of the disturbances. This effect has been discussed in the
Fabbiane et al., 2015; Sturzebecher and Nitsche, 2003). This section literature before (Fabbiane et al., 2015; Pätzold, 2013). The following
discusses the stability of the control algorithm for changes in UWT results show a detailed analysis of the controller stability based on
based on the investigation on the boundary-layer transmission be- the phase-angle error, which is caused by the changing phase speed
havior in Section 3.2. An example for wave attenuation at UW T = of Hec . The phase-angle error due to changing free-stream velocity
13 m/s is shown in Fig. 16. The TS-wave “hump”, which ranges from e is shown in Fig. 18 for the frequency band of 50–400 Hz, rel-
about 120 Hz to 300 Hz, is attenuated almost completely. The nar- ative to the phase angle of the reference case at UW T = 12 m/s. Af-
row peaks at 150 Hz, 250 Hz and 350 Hz occur due to electromag- ter a (counter-)wave is excited by the plasma actuator c it travels
netic interference caused by the wind-tunnel motor power supply. downstream with the flow and is measured by the error sensor e.
The maximum disturbance attenuation Emax = 13 dB is achieved in a The phase-angle error e (f), relative to a reference phase angle
small band, representing the most amplified TS-wave frequencies. In ref (f), depends on the traveled distance xe − xc = l = 0.045 m, the
addition, disturbances below the TS-wave band are damped by the phase speed c and its frequency f = T1 :
plasma actuator. The amplitudes above 300 Hz seem to be amplified,
360◦ · l
which can be caused either by the turbulence production of the PA or e ( f ) = − re f ( f ). (6)
the electromagnetic interference caused by the plasma discharge. c·T
If the control algorithm is operated off the design point, at which The controller takes this phase-angle error into account by using
the model of the secondary path Ĥec has been extracted, the con- the model Ĥec , which is determined for a given situation (here a given
troller can become unstable due to the change of the transmission UWT ). If one of these parameters changes, the controller uses a wrong
behavior of the boundary-layer flow (Section 3.2). Fig. 17 shows the model which results in a phase-angle error. Using the LST the effect
maximum achieved reduction of the disturbance Emax as a function of varying free-stream velocities on the phase-angle error is investi-
of the wind-tunnel velocity during the experiment UWT and the ve- gated and compared to the experimental results. The experimental
locity UWT, ident where the secondary path model Ĥec,UW T was ac- and theoretically calculated e presented in Fig. 18 show differ-
quired. The investigation of the stability of the controller was con- ences but the qualitative behavior is basically the same. The reasons
ducted for different wind-tunnel velocities in steps of UW T = 1 m/s. for the deviation between LST (dashed curves) and experiment (solid
The white areas represent unstable controller operation (no cancela- curves) have been discussed in Section 3.2.
B. Simon et al. / International Journal of Heat and Fluid Flow 56 (2015) 344–354 351
Fig. 20. fxLMS kernels for controller operation at design point and off design opera-
tion.
Fig. 19. Stability of the fxLMS control algorithm with constant Ĥec,UW T at different UWT .
The absolute value of the phase-angle error e increases for de-
creasing flow velocity. The horizontal spacing of the curves decreases
for increasing UW T. On the other hand higher frequencies also cause a
higher phase-angle error and the most amplified frequency is higher
for increasing velocity.
For the phase-angle based investigations on the controller stabil-
ity the phase-angle error e is now calculated for each design point
UW T,ident = 7 − 17 m/s at which a model Ĥec,UW T has been identified.
The TS-wave bandpass is shifting to higher frequencies with higher
UWT and the phase-angle error e (f) is a function of frequency f.
Therefore the phase-angle error at the upper TS-wave bandpass limit Fig. 21. Measured and scaled/stretched secondary path model Ĥec for different wind-
(cmp. Fig. 9) is defined as the critical phase-angle error. The exam- tunnel velocities based on a reference UW T,ref = 12 m/s.
ple for UW T = 12 m/s shown in Fig. 18 leads to a phase-angle error of
e = 75◦ at the upper frequency limit of 300 Hz, if the controller is
operated off design at UW T = 14 m/s. (e.g. Zhang et al., 2003). However, these methods are usually very
Fig. 19 shows the phase-angle error at that upper frequency limit application specific and significantly increase the complexity of the
caused by an off design controller operation. For each secondary path controller. In cases where the changes in the secondary path are dom-
model Ĥec,UW T , identified at UWT, ident , the wind-tunnel velocity UWT inated by a small number of dominant parameters, an online adapta-
has been varied from UW T = 7 − 17 m/s. The dashed lines connect the tion can be conducted using look up tables of secondary path esti-
data points which belong to the same UWT whereas the wind-tunnel mates for in parameter range of interest. The secondary path is then
speed rises from bottom to top in steps of UW T = 1 m/s. All data adopted online by measuring this dominating parameter and inter-
points marked with filled circles show stable fxLMS controller opera- polating within the look up table. For this approach a lot of effort is
tion and a successful wave attenuation, all other configurations went required for the initial experimental identification of the secondary
unstable (cmp. Fig. 17). It is commonly known that the fxLMS control path look up table. Alternatively, if there is a functional dependency
algorithm can theoretically compensate a ± 90° phase-angle error be- of the secondary path on the dominant parameters, then the sec-
tween the secondary path model Ĥec and the physical path Hec by ondary path model may be extrapolated from a single reference sec-
adapting the filter online (Hansen, 2002; Snyder and Hansen, 1994). ondary path model.
These two borders are drawn as solid horizontal lines. The postpro- Investigations of the transmission behavior of the boundary-layer
cessing based on the phase-angle error e demonstrates the flexi- with different flow speeds in Section 3 indicate that it is possible
bility of the adaptive control algorithm in the band of ± 90°. Almost adapt the secondary path model Hec by stretching and scaling the
all configurations in this band are stable. impulse-response model Ĥec with changing UWT (cmp. Fig. 13). This
The fxLMS control algorithm is able to adapt the filter w on- Section discusses the extension of the stable controller operation
line and to compensate the phase-angle error between the physical range by adapting the secondary path model Ĥec .
secondary path Hec and its model Ĥec,UW T . Fig. 20 shows the fxLMS
kernels, acquired at UW T = [11, 13, 15]m/s and additionally (dashed 5.1. Parameterized scaling and stretching of Ĥec
lines) two cases with off-design operation. The adaption process of
the filter was completed in all cases shown. The kernels of the off- The secondary path model Ĥec can be adapted during the ex-
design cases differ only slightly from the on-design cases, which periments by stretching and scaling the reference impulse response
shows the ability of the fxLMS to overcome the phase-angle error in Ĥec,ref . This approach is depicted in Fig. 21 where the solid curves
the adaption process. show the measured Ĥec , copied from Fig. 10. The zero crossing be-
tween the global minimum and maximum was chosen as a measure
5. Adaptive secondary path for the temporal stretching and the value of the global minimum was
chosen as a parameter for the amplitude scaling. The thereby de-
The fxLMS controller has shown a robust behavior in a certain op- rived scaling and stretching parameters and the 4th order polynomi-
eration range, corresponding to ± 90° phase-angle error (Section 4). als are plotted in Fig. 22. The dashed curves in Fig. 21 show stretched
In order to extend the operational range of the controller it is neces- and scaled impulse responses based on a Ĥec for UW T,ref = 12 m/s
sary to adapt the secondary path online (Elliott, 2000; Kuo and Mor- compared to the models Ĥec measured at each velocity individually.
gan, 1995). Several methods for the continuous online identification The presented procedure matches the shape of measured impulse
of secondary paths have been investigated within the last decades responses quite well. It is updated depending on the free-stream
352 B. Simon et al. / International Journal of Heat and Fluid Flow 56 (2015) 344–354
Fig. 22. Scaling and stretching parameters for secondary path model adaption. Fig. 24. fxLMS kernel for design point operation (solid curves) and adaption of Ĥec
with UW T,re f = 12 m/s (dashed curves).
Fig. 25. Surface hot-wire signals p1 (t) and p2 (t) for white noise excitation.
Fig. 23. Controller performance Emax with adaptive secondary path Ĥec , adaption
velocity. Towards the development of more capable boundary-layer
based on UWT with reference UW T,ref = 12 m/s.
control systems it seems beneficial to conduct the adjustment based
velocity during the controller operation in the experiment. This pro- on locally obtained information on the flow. Alternatively to the ex-
cedure allows a dynamic adaption to changing flow conditions. Nev- ternal measure UWT the secondary path Ĥec can also be adapted based
ertheless a calibration of the secondary path model adaption method on the estimate of the phase-angle difference d between the sig-
is unavoidable which means, that the impulse response models have nals captured at the two upstream sensors p1 and p2 (Fig. 25). The
to be recorded before controller operation. phase-angle difference d is directly related to UWT via the phase
speed c of the disturbances (Section 3.2). This technique allows to
5.2. Controller stability with adaptive secondary path Ĥec adapt Ĥec based on signals measured locally by wall mounted sensors
and does not rely on a remote measure of UWT . Therefore an inter-
The stability of the basic fxLMS control algorithm is based on Ĥec esting application for this technique are flight experiments, where
as a model of the boundary-layer transmission behavior. Because of not only the flow speed but also the pressure distribution can vary in
its ability to adapt, the basic control algorithm also works well with time. The phase-angle difference d catches the phase speed of the
slight changes of the flow velocity but becomes unstable for larger ve- disturbance by implication and therefore adapts the controller based
locity deviations (cmp. Fig. 17). The circles in Fig. 23 show the maxi- on the phase speed (propagation speed) of the disturbances. This is
mum disturbance attenuation Emax for constant secondary path mod- an important feature because the wave cancelation success mainly
els Ĥec determined for each wind tunnel velocity UWT and operated at depends on the correct phase angle relation between the disturbance
each design point. The solid line shows the attenuation obtained with and the counteracting force. The following paragraph describes the
a single secondary path model Ĥec adapted online as described in the implementation of the phase-angle difference detection algorithm,
previous section. Deviations in the control success Emax between both which is indicated with a P in the block diagram of the control algo-
cases at the same UWT result from difficulties with repeatable recep- rithm in Fig. 1.
tivity properties (Section 3.1) and are unlikely caused by a lower con- Fig. 25 shows the amplified AC signals of the surface hot-wires p1
trol success due to the adapted secondary path. The main benefit of and p2 at UW T = 12 m/s. The disturbances created upstream of the
the adaptive secondary path model Ĥec is a broader operation range sensors travel downstream with the phase speed c and cause a shift
while the fxLMS controller with constant Ĥec only can adapt phase d between both sensor signals. The short distance between both
changes up to 90°. sensors only lead to a limited phase-angle difference but it can be
As a comparison to the fxLMS kernels without secondary path extracted as a measure for the phase speed c and UWT , respectively.
adaption (Fig. 20) the kernels with adapted/stretched Ĥec are shown An algorithm for phase-angle difference detection has been imple-
in Fig. 24. In addition the kernel for design-point operation is plotted. mented in order to run it online on the dSPACE system. The data is
The shape in both cases is similar but the amplitude of the first peak first filtered with a narrow 8th order IIR bandpass filter at 200 Hz.
(local maximum) is lower. This underestimated amplitude of the first The resulting signal is then normalized in between one cycle dura-
peak does not seem to alter the stability properties of the controller tion T200 Hz = 0.005 s. The normalized signals p1, n and p2, n are now
and the control success Emax significantly. approximated as two sinusoidal signals with the same frequency f, as
indicated in Eqs. (7) and (8).
5.3. Embedded sensors for flow-state identification
p1,n = sin (2π f t + 1 ) (7)
The previously described approach for the adaption of the sec-
ondary path model is based on the knowledge of the free-stream p2,n = sin (2π f t + 2 ) (8)
B. Simon et al. / International Journal of Heat and Fluid Flow 56 (2015) 344–354 353
Fig. 27. Controller performance Emax with adaptive secondary path Ĥec . Compari-
son between adaption based on UWT and phase-angle difference d with reference
UW T,ref = 12 m/s.
Acknowledgments Kriegseis, J., Moeller, B., Grundmann, S., Tropea, C., 2011. Capacitance and power con-
sumption quantification of dielectric barrier discharge(DBD) plasma actuators. J.
Electrostat. 69 (4), 302–312.
This work was supported by the German research foundation Kriegseis, J., Schwarz, C., Tropea, C., Grundmann, S., 2013b. Velocity-information-based
(DFG) under the grant No. GR 3524/4-1. force-term estimation of dielectric-barrier discharge plasma actuators. J. Phys. D:
Appl. Phys. 46 (5), 055202.
References Kuo, S.M., Morgan, D., 1995. Active Noise Control Systems: Algorithms and DSP Imple-
mentations. John Wiley & Sons, Inc.
Kurz, A., Goldin, N., King, R., Tropea, C., Grundmann, S., 2013. Hybrid transition control
Borodulin, V., Kachanov, Y.S., Koptsev, D., Roschektayev, A., 2002. Experimental study
approach for plasma actuators. Exp. Fluids 54 (11), 1–4.
of resonant interactions of instability waves in self-similar boundary layer with an
Kurz, A., Simon, B., Tropea, C., Grundmann, S., 2014. Active wave cancellation using
adverse pressure gradient: Ii. detuned resonances∗. J. Turbul. 3 (062).
plasma acutators in flight. In: Proceedings of the 52st AIAA Aerospace Sciences
Cattafesta, L.N., Sheplak, M., 2011. Actuators for active flow control. Annu. Rev. Fluid
Meeting.
Mech. 43, 247–272.
Moreau, E., 2007. Airflow control by non-thermal plasma actuators. J. Phys. D: Appl.
Corke, T.C., Enloe, C.L., Wilkinson, S.P., 2010. Dielectric barrier discharge plasma actua-
Phys. 40, 605–636.
tors for flow control. Ann. Rev. Fluid Mech. 42 (1), 505–529.
Pätzold, A., 2013. Experimentelle Untersuchungen aktiv ausgelenkter Oberflächen zur
Duchmann, A., Reeh, A., Quadros, R., Kriegseis, J., Tropea, C., 2010. Linear stability anal-
Reduktion der Wandreibung durch Transitionsverzögerung. TU Berlin Ph.D. thesis.
ysis for manipulated boundary-layer flows using plasma actuators. In: Proceed-
Peltzer, I., Pätzold, A., Nitsche, W., 2009. In-flight experiments for delaying laminar-
ings of the Seventh IUTAM Symposium on Laminar-Turbulent Transition. Springer,
turbulent transition on a laminar wing glove. Proc. Inst. Mech. Eng., Part G: J.
pp. 153–158.
Aerosp. Eng. 223 (6), 619–626.
Duchmann, A., Simon, B., Tropea, C., Grundmann, S., 2014. Dielectric barrier discharge
Reeh, A.D., 2014. Natural Laminar Flow Airfoil Behavior in Cruise Flight through Atmo-
plasma actuators for in-flight transition delay. AIAA J 52 (2), 358–367.
spheric Turbulence. TU Darmstadt Ph.D. thesis.
Elliott, S., 2000. Signal Processing for Active Control. Academic press.
Saric, W., Reed, H., Kerschen, E., 2002. Boundary-layer receptivity to freestream distur-
Fabbiane, N., Semeraro, O., Bagheri, S., Henningson, D.S., 2014. Adaptive and model-
bances. Annu. Rev. Fluid Mech. 34 (1), 291–319.
based control theory applied to convectively unstable flows. Appl. Mech. Rev. 66
Snyder, S.D., Hansen, C.H., 1994. The effect of transfer function estimation errors on the
(6), 060801. doi:10.1115/1.4027483.
filtered-x lms algorithm. IEEE Trans. Signal Process. 42 (4), 950–953.
Fabbiane, N., Simon, B., Fischer, F., Grundmann, S., Bagheri, S., Henningson, D.S., 2015.
Sturzebecher, D., Nitsche, W., 2003. Active cancellation of Tollmien–Schlichting insta-
On the role of adaptivity for robust laminar flow control. J. Fluid Mech. 767.
bilities on a wing using multi-channel sensor actuator systems. Int. J. Heat Fluid
Grundmann, S., Tropea, C., 2008. Active cancellation of artificially introduced Tollmien–
Flow 24 (4), 572–583.
Schlichting waves using plasma actuators. Exp. Fluids 44 (5), 795–806.
Thomas, A.S., 1983. The control of boundary-layer transition using a wave-
Hansen, C.N., 2002. Understanding Active Noise Cancellation. CRC Press.
superposition principle. J. Fluid Mech. 137, 233–250.
Joslin, R., 1998. Aircraft laminar flow control 1. Annu. Rev. Fluid Mech. 30 (1), 1–29.
Würz, W., Sartorius, D., Kloker, M., Borodulin, V., Kachanov, Y., Smorodsky, B., 2012.
Kotsonis, M., Giepman, R., Hulshoff, S., Veldhuis, L., 2013. Numerical study of the con-
Nonlinear instabilities of a non-self-similar boundary layer on an airfoil: experi-
trol of Tollmien–Schlichting waves using plasma actuators. AIAA J. 51 (10), 2353–
ments, dns, and theory. Eur. J. Mech.-B/Fluids 31, 102–128.
2364.
Zhang, M., Lan, H., Ser, W., 2003. A robust online secondary path modeling method with
Kriegseis, J., Duchmann, A., Tropea, C., Grundmann, S., 2013a. On the classification of
auxiliary noise power scheduling strategy and norm constraint manipulation. IEEE
dielectric barrier discharge plasma actuators: a comprehensive performance eval-
Trans. Speech Audio Process. 11 (1), 45–53.
uation study. J. Appl. Phys. 114 (5), 053301.