An artificial neural network model applied to convert sucrose chord length distributions into particle size distributions

Autor(es)

Crestani, C. E. ^[1] ; Bernardo, A. ^[2] ; Costa, C. B. B. ^[3] ; Giulietti, M. ^[2]

Afiliação do(s) autor(es)

^[1] Fed Inst Educ Sci & Technol Sao Paulo, Rua Stefano DAvassi 625, BR-15991502 Matao, SP - Brazil

^[2] Univ Fed Sao Carlos, Dept Chem Engn, Rodovia Washington Luis, Km 235, BR-13565905 Sao Carlos, SP - Brazil

^[3] Univ Estadual Maringa, Dept Chem Engn, Ave Colombo 5790, Bloco D90, BR-87020900 Maringa, Parana - Brazil

• Tipo de documento: Artigo Científico
• Fonte: Powder Technology; v. 384, p. 186-194, MAY 2021.
• Citações Web of Science: 0

Resumo

Online monitoring of the solid phase in industrial processes of sugar crystallization is still a challenge. Laser back scattering is one of the most promising techniques; however, the measured chord length distribution (CLD) does not have a physical meaning of crystal size. This work converted sucrose CLD measured by an online sensor into particle size distribution (PSD) using an artificial neural network (ANN). CLD and suspension concentration of 116 experiments were the input to the ANN and PSD was its output. The trained ANN exhibited a coefficient of variation between experimental and calculated PSD of 0.998. Data of experimental sucrose crystallization was used to validate the model, resulting in a maximum deviation of 0.090 mm in mean size and 6.16% in the coefficient of variation of distribution. This model may be used to improve both industrial processes (process optimization and control) and laboratory studies (kinetics determination). (c) 2021 Elsevier B.V. All rights reserved. Online monitoring of the solid phase in industrial processes of sugar crystallization is still a challenge. Laser backscattering is one of the most promising techniques; however, the measured chord length distribution (CLD) does not have a physical meaning of crystal size. This work converted sucrose CLD measured by an online sensor into particle size distribution (PSD) using an artificial neural network (ANN). CLD and suspension concentration of 116 experiments were the input to the ANN and PSD was its output. The trained ANN exhibited a coefficient of variation between experimental and calculated PSD of 0.998. Data of experimental sucrose crystallization was used to validate the model, resulting in a maximum deviation of 0.090 mm in mean size and 6.16% in the coefficient of variation of distribution. This model may be used to improve both industrial processes (process optimization and control) and laboratory studies (kinetics determination). (AU)

• Processo FAPESP: 11/51902-9 - Simulação da biorrefinaria de cana-de-açúcar de 1ª. geração na plataforma EMSO
• Beneficiário: Antonio Maria Francisco Luiz Jose Bonomi
• Modalidade de apoio: Auxílio à Pesquisa - Programa BIOEN - Temático