Biomass estimation using artificial neural networks on field programmable analog devices

In biotechnological processes it is always desired to produce biomass and/or a secondary product. There are many parameters related to these variables, such as pH, dissolved oxygen and many others. Most variables can be measured on line, nevertheless, there are variables that cannot be measured on-line such as biomass and secondary products in fermentation. Knowledge of the values of these variables is essential, because at least one of them is the goal of any fermentation. Artificial neural networks (ANN) are used to estimate variables that are difficult or costly to measure using other process variables that are more readily available. Our work is based on the study, design, implementation and testing of a solution based on ANN in mainly analog hardware. Ours is a proposed solution to the problem of estimating biomass and product concentrations in fermentation producing the Astaxantin pigment. In this work we present the results of the implementation in analog reconfigurable hardware of an ANN solution for the estimation of biomass in a batch fermentation process. The ANN was designed using MATLAB. Its architecture was the multilayered perceptron, the hidden and output layers use nonlinear functions. We use TRAC (totally reconfigurable, field programmable analog devices) analog hardware for our implementation of the networks. Our aim is, at the end of this project, to have a hardware version of the ANN estimators working on-line and providing information about the biomass or secondary product concentration