Uso de técnica de espectroscopia NIR no desenvolvimento de softsensor para monitoramento on-line da fermentação alcoólica industrial

• Ano de defesa: 2024
• Autor(a) principal: Aidar Neto, Homero
• Orientador(a): Ribeiro, Marcelo Perencin de Arruda
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Química - PPGEQ
• Palavras-chave em Português: Fermentação alcoólica; Monitoramento; Espectroscopia NIR; Softsensor
• Palavras-chave em Inglês: Alcoholic fermentation; Monitoring; NIR spectroscopy
• Área do conhecimento CNPq: ENGENHARIAS::ENGENHARIA QUIMICA::PROCESSOS INDUSTRIAIS DE ENGENHARIA QUIMICA
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/20200

Automated control and monitoring of bioprocesses are fundamental tools for the modern biotechnology industry. Chemical introduction, for example, is a biotechnological process already well established in the sugar and alcohol industry, with high productivity and robust microorganisms. However, one factor can still be improved: bioprocess monito- ring. Detailed fermentation monitoring is carried out with at-line and off-line techniques, such as High Performance Liquid Chromatography (HPLC), which implies a delay to verify the end of fermentation and/or some disturb. NIR spectroscopy can be used to monitor bioprocesses, inferring key variables in fermentation in real time, such as resi- dual sugar concentration. A major difficulty in using this technique is the presence of intense noise at frequencies, requiring pre-treatment for its use in a soft sensor. The use of a phenomenological interference model can reduce the noise associated with experi- mental data, improving the estimation of interference variations using the chemometric technique. Given these aspects, a methodology was developed for inferring cell concentra- tion, substrate concentration and product concentration in a prepared fermentation. The methodology is based on smoothing experimental data from samples using a fermentative kinetic model. Such data were correlated with NIR spectra to adjust the chemometric model PLSR (Partial Least Squares Regression) and construct the virtual sensor. Three fed-batch fermentations were used to construct the detection set and three batches were used to validate the sensor.