Modelling fireside corrosion of superheaters and reheaters following coal and biomass combustion.

Specific data analysis methods Principal Component Analysis (PCA) and Partial Least Squares (PLS) have been employed to develop a novel series of fireside corrosion models showing different rates for coal vs biomass, and the UK, US and world-traded coals with different controlling factors. The results from the analysis reveal the potential influence of corrosion resistant alloying elements and fuel contaminants on the fireside corrosion rates which are subsequently employed for fireside corrosion model building. Achieving this aim has involved two key steps i.e. (1) Fireside corrosion data from various plant investigations have been collated for this modelling work, (2) Fireside corrosion model development using PCA and PLS. The results from UK coal fireside corrosion data indicate that the fireside corrosion rate, while requiring the presence of S, is dependent on the fuel’s Cl, Na, and K levels as S is present in excess. The results from the UK woody biomass-fired data show the fireside corrosion rate dependent mainly on K, Ca, and Cl levels from the fuel. Co-firing of coal/biomass highlights slightly different fuel species including Cl, K, Mg, S and P. Data subsets based on probe investigations in the US following coal combustion depends on different variables for modelling when compared to UK coals and thus the fireside corrosion damage is attributed to mainly the fuel's chemistry. High performances of prediction models featuring austenitic tubes under coal/biomass, and ferritic/nickel-based tubes under biomass firing were achieved. However, the predictive performances of the ferritic and nickel-based metals under coal have been the least successful. The PLS method is best suited for prediction as it uses the input data (alloy/fuel elements, metal temperature) to find the most critical variables that have maximum covariance with the dependent variable; i.e. the corrosion rate, as opposed to PCA which only maximises variability in the input data.