Artificial neural networks as a biomass virtual sensor for a batch process

The ability of artificial neural networks (ANN) to learn from experience rather than from mechanistic descriptions is making them the preferred choice to model processes with intricate variable interrelations. We apply ANN as a data fusion method to provide estimations of biomass in a batch fermentation process. The readings of biomass must be periodic, of the desired frequency and reliable to a 5% error. A desired feature is that the measurement method be robust to sensor perturbations and failures. The robustness of the presented estimator system has been tested with simulated noisy inputs and with sensor failures and a mean average error of near 5% has been obtained. A new technique is presented as a data fusion method. The technique is tested on real process data. Simulated tests are applied to evaluate performance and robustness. We suggest that an artificial neural network may be used to obtain an insight on the relative influence of each of the variables used at every stage of the process.