EUBCE 2026 - Vittor ALVES - Multivariate and Machine Learning Analysis of Biomass Combustion Regimes in a Bubbling Fluidized Bed

Multivariate and Machine Learning Analysis of Biomass Combustion Regimes in a Bubbling Fluidized Bed

Short Introductive summary

Biomass combustion remains a key pathway for renewable energy generation, yet operational challenges linked to ash, chlorine, sulphur, and nitrogen hinder its large-scale application. This study introduces a data-driven framework combining Principal Component Analysis (PCA), k-means clustering, and Linear Discriminant Analysis (LDA) to classify combustion and emission behaviors of diverse biomasses—such as elephant grass, palm oil residual biomass, and their leached and blended forms—burned in a bubbling fluidized bed combustor. By integrating multivariate statistics, the approach identifies stable and emissive operational regimes, correlating them with physicochemical properties and process variables. The results reveal that leaching significantly reduces SO2 and NO? emissions, enhancing combustion stability and promoting cleaner regimes, while mineral-rich biomasses shift toward higher-emission patterns. The PCA–k-means–LDA pipeline proves to be a powerful tool for diagnosing combustion dynamics, supporting predictive emission control, and guiding the design of low-emission bioenergy systems aligned with sustainable energy transition goals.

V. Alves, Instituto de Pesquisas Tecnológicas de São Paulo, BRAZIL
A.H. Ushima, Instituto de Pesquisas Tecnológicas de São Paulo, BRAZIL
D.C. Meyer, Instituto de Pesquisas Tecnológicas de São Paulo, BRAZIL
B.V.R. Bezerra, Brasil Biomass, São João da Baliza, BRAZIL