Main contents

Multiparameter polynomial models for monitoring and management of fermentation and food processes

The project belongs to a larger Dynamic Systems Modeling (DYSMO) project coordinated by VTT Automation. DYSMO is one of the projects of Adaptive and Intelligent Systems Applications 1994-1999 -program of TEKES.

Project description

The primary goal was to develop model based solutions for monitoring and management of fermentation processes producing enzymes and for chromatographic separation processes. Both processes are difficult to model with traditional methods. These processes were selected to be used as testbeds for developing dynamic NN-models based on Volterra and Wiener theories for nonlinear dynamic systems.

The original Wiener representation consists of linear dynamics (Laguerre filters) and static nonlinear mapping (polynomial expansion). In Wiener-NN, static nonlinear mapping is approximated with NN. The feedforward Wiener-MLP is suitable for modeling of finite memory systems. The Wiener-MLP with feedback can also be used for modeling of autonomous type systems, like fermentation processes.

Due to robust orthogonal (e.g. Laguerre) description of signals, Wiener-MLP or -SOM classifiers are advantageous for recognizing of spatio-temporal patterns needed in process monitoring. A method using Wiener-type MLP or SOM classifiers for detecting and recognizing the functional states on-line was developed. It was demonstrated with Bacillus subtilis fermentation.

The Laguerre descriptions of the signals realize the state in different Wiener-NNs. In the NOE case, the state-space formulation was utilized in parameter estimation with the Extended Kalman filter, which can be also used directly as a state estimator in connection with the NFIR and NOE-type Wiener-NN models.

The various Wiener-models were applied to model the dynamics of industrial chromatographic separation columns and industrial Tricoderma fungi fermentation for monitoring, state estimation, on-line simulation and fault diagnosis. Chromatographic separation process was also modeled with hybrid models: some parameter dependencies in the traditional mechanistic model (partial differential equation) were described with different black-box mappings.

More information


(S) File not found or open failed result: raw file: /sovellus/trip_htdocs/tkk/julkaisee/raw.*