THESE de DOCTORAT - cerfacs

THESE de DOCTORAT - cerfacs THESE de DOCTORAT - cerfacs

22.10.2014 Views

92 Chapter 5: Assessment of combustion noise in a premixed swirled combustor after extracting hydrodynamics is placed relatively far from that one predicted by the hybrid approach, as can be observed from Fig. 5.25. Pressure Fluctuation (Pa) M5 M6 M7 40 Premixer Combustion chamber 20 0 −20 −40 −0.2 −0.1 0 0.1 0.2 x (m) 0.3 0.4 0.5 (a) Pressure Wave from Eq. (2.63) Hybrid Computation Pressure Fluctuation (Pa) M5 M6 M7 40 Premixer Combustion chamber 20 0 −20 −40 −0.2 −0.1 0 0.1 0.2 x (m) 0.3 0.4 0.5 (b) Pressure fluctuation from LES after filtering Direct Computation Figure 5.25: Longitudinal pressure Waves oscillating at 954 Hz Pressure Fluctuation (Pa) M5 M6 M7 40 Premixer Combustion chamber 20 0 −20 −40 −0.2 −0.1 0 0.1 0.2 x (m) 0.3 0.4 0.5 (a) Pressure Wave from Eq. (2.63) Hybrid Computation Pressure Fluctuation (Pa) M5 M6 M7 40 Premixer Combustion chamber 20 0 −20 −40 −0.2 −0.1 0 0.1 0.2 x (m) 0.3 0.4 0.5 (b) Pressure fluctuation from LES after filtering Direct Computation Figure 5.26: Longitudinal pressure Waves oscillating at 1658 Hz Considering now the acoustic energy, it is observed from Figs. 5.27 and 5.28 that the values predicted by the hybrid approach are in a good agreement to those computed by LES after extracting the acoustic content from the complete pressure fluctuations. It is noticeable then that the energy coming from hydrodynamic fluctuations has been removed. Note that some caution must be taken when computing the Fourier transform of the divergence of the velocity (See T1 Eq. 5.20). If a rectangular window is applied to the temporal signal, a really good correspondance is seen only for frequencies lower than 400 Hz (not shown). On the other hand, if a Gaussian window is applied, a good agreement is seen over the entire frequency band, except at a very low frequencies. This is the case for the spectra shown (Figs. 5.21, 5.22, 5.27 and 5.28) where the signal does not capture the good acoustic level before 200 Hz.

5.6 Conclusions 93 Ac. Energy (J) − micro 5 Hybrid Computation Direct Computation: LES Direct Computation; LES (Filtered) 10 −5 0 500 1000 1500 2000 2500 10 0 Frequency (Hz) Ac. Energy (J) − micro 6 Hybrid Computation Direct Computation: LES Direct Computation; LES (Filtered) 10 −5 0 500 1000 1500 2000 2500 10 0 Frequency (Hz) Figure 5.27: Acoustic energy. Direct and hybrid approaches Ac. Energy (J)− micro 7 Hybrid Computation Direct Computation: LES Direct Computation; LES (Filtered) 10 −5 0 500 1000 1500 2000 2500 10 0 Frequency (Hz) Figure 5.28: Acoustic energy. Direct and hybrid approaches 5.6 Conclusions Combustion noise of a premixed swirled combustor has been assessed by two different numerical approaches: a direct computation, in which the noise produced by the flame is calculated together with the flow and flame dynamics, and a hybrid computation, in which the acoustic field is evaluated from the sources of noise in a separate step. Classical comparisons between mean and fluctuating (rms) velocity fields were performed between two LES on a coarse mesh (3 millions cells) and a refined mesh (10 millions cells), and PIV measurements. Mean velocity fields (axial and radial) were well predicted by both LES cases, whereas only the refined mesh succeed in recovering the proper rms velocity fields. It was then observed that satisfactorily predicting the velocity fluctuating field does not mean to reproduce correct flame dynamics and heat release. On the one hand, the mean heat release corresponding to the experimental thermal power is well captured by both LES. On the other hand, significative differences are found between the two simulations when looking at the shape of the instantaneous heat release and its rate of change. As a consequence, a correct estimate of the combustion noise radiation is not reached either. Several phenomena might be the cause of such a misprediction. A lack in numerical resolution can be one possible explanation: computing the small turbulent length scales in the shear flow region might be significant since these eddies might have a non-negligible influence on the flame dynamics and, as a con-

5.6 Conclusions 93<br />

Ac. Energy (J) − micro 5<br />

Hybrid Computation<br />

Direct Computation: LES<br />

Direct Computation; LES (Filtered)<br />

10 −5<br />

0 500 1000 1500 2000 2500<br />

10 0 Frequency (Hz)<br />

Ac. Energy (J) − micro 6<br />

Hybrid Computation<br />

Direct Computation: LES<br />

Direct Computation; LES (Filtered)<br />

10 −5<br />

0 500 1000 1500 2000 2500<br />

10 0 Frequency (Hz)<br />

Figure 5.27: Acoustic energy. Direct and hybrid approaches<br />

Ac. Energy (J)− micro 7<br />

Hybrid Computation<br />

Direct Computation: LES<br />

Direct Computation; LES (Filtered)<br />

10 −5<br />

0 500 1000 1500 2000 2500<br />

10 0 Frequency (Hz)<br />

Figure 5.28: Acoustic energy. Direct and hybrid approaches<br />

5.6 Conclusions<br />

Combustion noise of a premixed swirled combustor has been assessed by two different numerical<br />

approaches: a direct computation, in which the noise produced by the flame is calculated<br />

together with the flow and flame dynamics, and a hybrid computation, in which the acoustic<br />

field is evaluated from the sources of noise in a separate step.<br />

Classical comparisons between mean and fluctuating (rms) velocity fields were performed between<br />

two LES on a coarse mesh (3 millions cells) and a refined mesh (10 millions cells), and<br />

PIV measurements. Mean velocity fields (axial and radial) were well predicted by both LES<br />

cases, whereas only the refined mesh succeed in recovering the proper rms velocity fields. It<br />

was then observed that satisfactorily predicting the velocity fluctuating field does not mean<br />

to reproduce correct flame dynamics and heat release. On the one hand, the mean heat release<br />

corresponding to the experimental thermal power is well captured by both LES. On the<br />

other hand, significative differences are found between the two simulations when looking at<br />

the shape of the instantaneous heat release and its rate of change. As a consequence, a correct<br />

estimate of the combustion noise radiation is not reached either. Several phenomena might be<br />

the cause of such a misprediction. A lack in numerical resolution can be one possible explanation:<br />

computing the small turbulent length scales in the shear flow region might be significant<br />

since these eddies might have a non-negligible influence on the flame dynamics and, as a con-

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!