Process state classification of fed-batch fermentation based on process variables analysis

• New manual classification methodology of the fermentation process combining expert knowledge and the signal analysis.
• New automatic classification methodology based on statistical analysis.
• Tested on the real process of cultivation Pseudomonas putida KT2442.

The success of fermentation processes operated in the fed-batch mode depends, among other factors, on appropriate substrate feeding. Overfeeding should be avoided because even slightly higher concentrations may result in an inhibition or even poisoning of the microbial culture when toxic substrates are used. Therefore, bioprocess monitoring and control play a key role. This paper introduces a new bioprocess state classification methodology combining expert knowledge and automatic signal analysis, suitable for on-line application. A fed-batch cultivation of the strain Pseudomonas putida KT2442 grown on octanoic acid was used as a model process. The classification was performed in two steps – a manual classification done by an expert, and a subsequent automatic classification using the results obtained by the manual classification. The manual classification strategy was based on the analysis of time profiles of selected process variables related to substrate feeding, such as dissolved oxygen tension and substrate feeding rate. Three process states were recognized – normal feeding, overfeeding and underfeeding. Ridge regression was then applied to the data results of the manual classification in order to design an automatic classification strategy for easier on-line use. This strategy can distinguish between the normal and other feeding states, using a limited number of on-line measurable output variables and their time fluctuations.