Soft-Sensors for Monitoring B. Thuringiensis Bioproduction.

One of the main challenges in ferentation cultures is the monitoring of key variables that can indicate the performance and help in the optimization of the bioprocess. On-line estimation could be a challenging task when an accurate process model is not available. Support Vector Machine (SVM) is an attractive and relatively simple method that can be used as an alternative to predict key variables by using several physical parameters measured online. In this paper, we show the application of SVM to the production of protein by B. thuringiensis. The soft-sensor was trained and validated with independent data sets of batch fermentations evaluating the impact of different predictor variables and kernels. Results show that protein production can be predicted only using online measurements.