IN

1
IN
Flame Spectrum Deconvolution
F. Ferro1, J. Sanches2 and E. C.Fernandes1
1Laboratory of Thermofluids, Combustion and
Energy Systems - IN 2Institute for Systems and
Robotics Instituto Superior Técnico1049-001
Lisbon, Portugal

• Abstract
• In this work it is performed the deconvolution of
  the flame spectra in order to estimate the real
  behavior from the individual chemiluminescent
  emissions.
• Firstly, it was analyzed a laser light spectrum
  emissions in order to obtain the optical system
  Point Spread Function and finally it was used to
  deconvolved the flame spectrum.
• The spectral data used in this paper was obtained
  from laminar premixed hydrocarbon air flames with
  equivalence ratio range (0.8?1.25) for propane
  and (0.8?1.25) for methane air flames.

• Experimental Results
• Spectrometer Point Spread Function

Spectra PhysicsSP2010 -05 Laser Spectrum
Problem Formulation

• Flame Spectrum Deconvolution

Flame spectra can be described by a Gaussian
Mixture Model
M- Number of chemical species ?k- Wavelength of
kth species ?k- Spectra height of kth specie
?k2- Standard deviation
Laminar Premixed Propane Air Flame Spectrum ?1
and P0.37kW in a Bunsen burner type with 10 mm
inner diameter
Due to optical system characteristics each
chemiluminescent emission contribution to the
flame spectrum appear more broader that in
reality is.

• Applications- Equivalence Ratio Estimation

Method I Fundamental peak emission estimated
by the proposed method. Method II Maximum peak
emission subtracted to the background broadband
obtained by a quadratic fit interpolation. Method
III Maximum peak emission with no subtraction

• Propane Air Flames

Real Flame Spectra can be obtained by the
Convolution Theorem
Real Spectrum (Gaussian Mixtures)

• Methane Air Flames

Measured Spectrum (Gaussian Mixtures)

• Acknowledgments
• The authors gratefully acknowledge the support of
  the European Commission within 6th Framework
  Programe, through the Marie Curie RTN Project
  "AETHER", Contract No MRTN-CT2006035713.
• The author F. Ferro and this work has been also
  sponsored by FCT under grant PTDC/EME-MFE/68830/
  2006

RecPad2010 - 16th edition of the Portuguese
Conference on Pattern Recognition, UTAD
University, Vila Real city, October 29th