Calibração multivariada aplicada ao monitoramento da síntese enzimática de biodiesel etílico utilizando lipase solúvel

Abstract

The production of biodiesel requires a process of monitoring and modeling of its reaction that guides the control of the production. Therefore, a multivariate calibration model that interprets the data obtained from biodiesel synthesis tests and allows in-line monitoring of the reaction was developed. Using liquid chromatography, the contents of monoglycerides (MG), diglycerides (DG), triglycerides (TG) and ethyl ester (ES) were obtained, and using near infrared spectroscopy (NIR), spectrum scans were obtained, reference data that was applied at the entry of the model. The analysis of the collected samples was conducted in a leave one out cross-validation procedure and the complexity of the model (number of latent variables - LV’s) was optimized for each component with the root mean square error of cross-validation (RMSECV). The multilinear methods applied in the treatment of data were partial least squares (PLS) and successive projection analysis (SPA). The models developed in PLS had 4 LV's for MG, 3 for DG, 4 for TG and 4 for ES, with respective errors of 17; 15; 5.67 e 8.28%. The models developed in SPA had 7 LV's for MG, 8 for DG, 11 for TG and 6 for ES, with respective errors of 20; 17.1; 11.7 e 14.1%, showing a greater precision of the models developed in PLS. Moreover, the results obtained for TG and ES presented a better adaptation of the components to the proposed model, as they result in minor errors in both methodologies, making it possible to monitor the conversion of TG and the yield of ES. Therefore, the results show a good predictive capacity for TG and ES with errors around 7% for PLS and 13% for SPA in relation to the maximum concentration obtained from these analytes. In addition, the PLS method was more accurate as it resulted in minor errors from less complex models.