276965 Online Estimation of Glucose Concentration and Oil Contents in Microalgal Biodiesel Production System

Monday, October 29, 2012
Hall B (Convention Center )
Se-Kyu Oh1, Sung Jin Yoo1 and Jong Min Lee2, (1)School of chemical and biological engineering, Seoul National University, Seoul, South Korea, (2)School of Chemical and Biological Engineering, Seoul National University, Seoul, South Korea

Online estimation of glucose concentration and oil contents in microalgal biodiesel production system

Se-Kyu Oh, Sung Jin Yoo and Jong Min Lee*

School of Chemical and Biological Engineering, Seoul National University

Microalgae cultivation process has recently attracted much attention as a process for biodiesel production because of the potential of microalgae for lipid production. In this process, real-time measurements are important for the optimal control to maximize oil productivity. Unlike typical chemical processes, most of important variables are concentrations, which are difficult to measure. Pre-treatments and biological analysis equipments are required to measure the concentrations, and these tasks can take from 2 hours to 2 days. Although the concentrations are obtained offline, the measurements cannot be applied for feedback control because of the significant time delay between sampling and measurement.

In this study, we will present an integrated framework to predict glucose concentration and oil contents using Raman spectroscopy without any knowledge of spectroscopic analysis and biology. Raman spectroscopy is suitable for in-situ on-line monitoring in bioprocess. Raman spectrum has weak water interference peak and sample preparation is not required. In addition, sample or microalgae culture in bioreactor is not destroyed by Raman laser. Spectra are taken from microalgae culture using Raman spectroscopy, and Savitzky-Golay smoothing filter is used to reduce noise effect of the spectra. Fluorescence effects that inevitably exist in biological spectra are removed by Rolling-Circle Filter (RCF). Another challenge is that underlying relationship including linearity in correlations between the prediction variables and spectra is difficult to know. Hence, concentrations are estimated by multivariate analysis. This study compares different multivariate tools including Principal Component Regression (PCR), Partial Least Squares (PLS), Kernel PLS, Wide-kernel PLS and Support Vector Regression (SVR) to evaluate prediction performance and proposes the most suitable technique for online estimation applications.

Extended Abstract: File Not Uploaded