Search

An infrared sensor was used for measurements of spatial distribution of hydrocarbon concentration in a model of a gas turbine combustor, using absorption tomography along multiple lines of sight (LOS). This sensor has the potential for monitoring the degree of premixedness of reacting fuel and air in stationary gas turbine combustors, where operation with lean premixed mixtures is important for reduction of NOx emissions. A numerical simulation, using a stochastic representation of a turbulent hydrocarbon distribution in the cross section of a duct was generated, in order to evaluate the ability of computed tomography (CT) to reconstruct the initial distribution. The simulation considered the experimental execution of the measurements, the optimal number of parallel LOS and projection angles and the optimal reconstructed grid size that would be required to minimize the error of reconstruction of the distribution with the algebraic reconstruction tomography algorithm for planar parallel geometry of LOS. We conclude that the quality of reconstruction is increasing asymptotically as both parallel and angular projections increase. As the reconstructed object grid is refined and given a maximum acceptable error of reconstruction, there is a minimum threshold for both the amount of view angles and parallel projections. For instance, the spatially integrated error of reconstruction can remain below 30% in a 11 × 11 object grid for at least 10 parallel LOS and at least 6 view angles. For similar quality of reconstruction and an increasing number of projection angles, the required amount of parallel LOS is slightly decreasing. An experimental assessment of the spatial resolution of the sensor is presented by comparing infrared absorption (INFRA) measurements with those from optical planar visualisation of fuel concentration, based on a Mie scattering technique. The flow investigated was a confined coaxial flow of a 8 mm-diameter central premixed air/methane jet (initial equivalence ratio of 2; Reynolds number of 3800), surrounded by a co-flow of air (0.5 m/s) in a 144 mm diameter duct. The INFRA sensor combined with CT used in this case an 11 × 11 reconstruction grid with 25 parallel LOS and 6 projection angles, because of limitations particular to this experiment, and assumed an axisymmetric flow to provide information to improve the accuracy of the reconstruction. The numerical simulation predicted for this geometry an error of about 30%, which is lowered by the axisymmetric assumption. The CT concentration measurements for low radial gradients (i.e. for Z≫2.75D) had an error in the reconstructed hydrocarbon concentration profile (here calculated by integration over the full width of the duct), of 25% of the corresponding Mie scattering measurements, which were previously calibrated with a flame ionisation detector. The error was found to increase with the local radial gradient of the concentration, which was confirmed by a supplementary numerical simulation that investigated the influence of the spatial gradient of concentration and laser beam diameter. This computational work showed that INFRA combined with CT should have a maximum laser beam width of 20% of the reconstruction grid spatial resolution for optimal accuracy and that the technique can resolve gradient scales of half a pixel of the reconstruction grid size.