Data fusion and artificial neural networks for biomass estimation
The ability of artificial neural networks (ANNs) to learn from experience rather than from mechanistic descriptions makes them the preferred choice to model processes with intricate variable interrelations. Some of these processes can be found in the area of biotechnology. The authors aim to use ANNs and data fusion to provide better instrumentation for a fermentation process and eventually optimise its performance. Of particular interest is the robust estimation of biomass in the production of an antibiotic. Several feed-forward backpropagation neural networks (BPNs) have been chosen for the experiments using the Levenberg–Marquardt learning algorithm. Work has been carried out to test the generalisation capabilities and performance in the presence of noise and sensor failure. It has been observed that, given the appropriate training, data fusion and ANN methodology lead to estimation of these parameters with an accuracy comparable to instrumentation errors.
