Mechanisms and Mitigation of Agglomeration During Fluidized Bed Combustion of Biomass

Fluidized bed combustion technology is increasingly used for biomass fuels, due to the high variability of their energy density and composition. However, this technology is still susceptible to ash-related issues. Agglomeration is caused by ash melting onto or reacting with bed material to form alkali silicate melts, allowing bed particles to adhere together. Accumulation of agglomerates causes bed defluidization, and consequently unscheduled downtime. This thesis investigates agglomeration mechanisms and mitigation measures at the pilot-scale, focusing on agricultural fuels that have received less attention in literature and may be of interest for boiler operators. When burning wheat straw, the magnesium-iron silicate bed material olivine lengthened defluidization times versus silica sand, though this was not sufficient to make the fuel viable. The additives kaolin and dolomite prevented bed defluidization entirely when burning miscanthus, but had no effect with wheat straw, despite chemically reacting with both fuel ashes. In combination with thermochemical modelling, it was proposed that the poor breakdown of wheat straw pellet sand release of ash to their surface allows the pellet to act as a seed for agglomerate formation, hence additives proving ineffective. Agglomeration mechanisms were studied with different fuels, bed materials and additives. This included a novel analysis of agglomerates from different bed locations, and a spatially defined study of agglomerates from tests with additives, both of which revealed mechanisms in greater detail than previously reported. A novel thermochemical modelling approach using FactSage was applied to agglomerate compositional data, together with an appraisal of the software for agglomeration studies. Through collaboration with project sponsor Sembcorp Energy UK on their “Wilton 10” bubbling fluidized bed boiler, a 5-year fuel data set was studied to determine fuel quality improvement potential. Several analytical methods were applied, including a machine learning algorithm. Recommendations were made regarding fuel quality and sampling.