10th International Aerosol Conference September 2 - September 7, 2018 America's Center Convention Complex St. Louis, Missouri, USA
Abstract View
Deconvolution of Nanoparticle Size Distributions Measured in Combustion Processes
HARTMUT MÄTZING, Werner Baumann, Andrei Bologa, Alexandra Loukou, Nadine Teuscher, Petros Vlavakis, Hans-Joachim Gehrmann, Hanns Rudolf Paur, Dimosthenis Trimis, Dieter Stapf, KIT, Karlsruhe, Germany
Abstract Number: 692 Working Group: Combustion
Abstract Nanoparticles are frequent by-products in laboratory and industrial scale combustion processes. They may originate from soot formation, from nucleation/condensation of organic material and of evaporated ash, from break-up of agglomerates or just from incompletely burnt material (dust). Depending on conditions, the particle size distributions may exhibit several peaks, e.g. a nucleation and several accumulation modes. Often, these modes overlap and may be difficult to detect by visual inspection. Therefore, numerical tools are required for peak deconvolution. Recently, differential evolution (DE) has been recommended as suitable tool to deconvolute particle size distributions in the μm size range (Alderliesten, 2016). Here, the application of DE to nanoparticle size distributions is demonstrated.
The nonlinear least squares regression code (NLINLS) provided by Mishra (2007) was upgraded from F77 to Fortran 95/2003 and used to generate Fit4LogNorms, an interactive code for fitting up to nine log normal particle size distributions (PSD) to a given dataset. Inputs are the particle size dp [nm or μm], the number density, dN/dlog dp [cm-3 or m-3], and some control parameters. The code approximates the user specified number of peaks to the given size distribution and reports the mean particle size, the total particle number density and the distribution width for each peak together with the fit quality. Occasional peak identities are checked and detected according to user specified criteria like minimum difference of mean particle size or size distribution width. The code was tested against literature results (Alderliesten, 2016) and other test cases.
Fit4LogNorms was applied to measurements of selected laboratory and bench scale experiments:
(a) metal oxide particles/agglomerates in flat and conical premixed C2H4 and C3H8/air flames (Teuscher et al., 2016) (b) soot formation in flat, premixed, superadiabatic CH4/O2 flames (Sentko et al., 2016) (c) fly ash particles from a 100 kW wood chips boiler (Bologa et al., 2013) (d) fly ash particles at a bench scale, 2.5 MW wood dust burner (Baumann et al., 2017).
All inspected size distributions were found to be multimodal, containing up to 5 modes. Note in particular that valid extrapolations are applicable to size distributions which are incomplete because of experimental limitations or for other reasons.
These examples show that in lab as well as bench scale applications aerosol dynamics is very complex. The multiple mode size distributions may be due to simultaneous nucleation, condensation, coagulation and heterogeneous reactions. In addition, the particle sampling process itself may affect the size distribution, especially under highly reactive conditions. Peak deconvolution can deliver key information for a full understanding of the aerosol formation and behavior.
References 1. Alderliesten, M. (2016). Part. Part. Syst. Charact., 33, 675–697. 2. Bologa, A. et al. (2013). EUBCE, Copenhagen. 3. Baumann, W. et al. (2017). Energy Procedia, 120, 705–712. 4. Mishra, S.K. (2007). NLINLS (F77 DE code). http://mpra.ub.uni-muenchen.de/4949/. 5. Sentko, M. et al. (2016). 36th Int. Symp. Combustion, poster 4P032, Seoul. 6. Teuscher, N. et al. (2016). EAC 2016, Tours.