Modeling and experimental approaches in biomass pyrolysis: Artificial neural network application & guard-bed reactor integration

In this thesis, biomass pyrolysis was investigated from the perspectives of slow and fast pyrolysis processes to discuss different aspects. Three different biomass feedstocks which are refuse derived fuel (RDF), olive oil residue (OOR) and lignocellulosic forest residue (LFR) were selected as biomass sources. The first part of the study is based on slow pyrolysis studies to understand the thermal decomposition behaviors of the selected biomasses at high temperature region. In addition, kinetic behaviors of biomass pyrolysis were also modeled using model-fitting and model-free approaches. As a result of thermal experiments, it was concluded that the pyrolytic behaviors were mainly divided into three stages which are i) moisture removal, ii) main decomposition of the structure (hemicellulose and cellulose degradation together with the start of the lignin degradation) iii) lignin decomposition as well as gasification of fixed carbon, and decomposition of tars. The structural differences directly affect the pyrolytic behaviors under slow pyrolysis conditions. LFR and OOR had all of these three decomposition stages clearly. However, according to TG and DTG curves of RDF, thermal decomposition of RDF involved multiple stages between 190-890C that were comprised of decomposition of cellulose, hemicellulose and lignin as well as the plastics in the structure. It is already known that thermal and kinetic analysis of biomass decomposition have been studied for many years, as a result, there are numbers of studies in the literature. Therefore, to bring a new perspective to this field, an alternative modeling approach, Artificial Neural Network (ANN), was introduced in biomass decomposition within the scope of this thesis. The ANN application in biomass pyrolysis were discussed from two aspects. In the first step, an ANN model was created that can predict the pyrolytic behaviors of RDF, which is more heterogeneous than other two biomasses without the necessity of experimentation. At the further step, the prediction performance of ANN was tested for more than one biomass (LFR and OOR) and the data predicted by ANN model was used to calculate the pyrolytic activation energies of biomass pyrolysis. From the modeling studies, it can be concluded that ANN is a promising tool which can be used to reduce the number of the required experiments. Moreover, ANN can even help making overall thermal data estimation for the heating rates that are difficult to reach due to the physical limitation of thermal analyzer, very small amounts of feedstocks which does not let repetition experiments, different types of mixtures (biomass-coal or biomass-plastic) using their individual thermal data. In addition, integration of the guard bed reactor was studied. In the light of experimental results, it was concluded that integration of the guard bed played an important role to improve the bio-oil quality.