María C. Palancar

NEURAL NETWORK MODEL FOR FLUIDISED BED DRYERS

The application of an artificial neural network (ANN) to model a continuous fluidised bed dryer is explored. The ANN predicts the moisture and temperature of the output solid. A three-layer network with sigmoid transfer function is used. The ANN learning is made by using a set of data that were obtained by simulating the operation by a classical model of dryer. The number of hidden nodes, learning coefficient, size of learning data set and number of iterations in the learning of the ANN were optimised. The optimal ANN has five input nodes and six hidden nodes. It is able to predict, with an error less than 10%, the moisture and temperature of the output dried solid in a small pilot plant that can treat up to 5 kg/h of wet alpeorujo. This is a wet solid waste that is generated in the two-phase decanters used to obtain olive oil.

Fluidized Bed Drying of 2-Phase Olive Oil Mill By-Products

This study is focused on the fluidized bed drying of solid olive oil mill by-products (SOB) resulting from the 2-phase oil extraction method. The most usual method used in SOB drying is the rotary dryer, which revealed some problems and disadvantages that can be easily solved with the development of new drying techniques based on exhaustive studies of the solid and a better comprehension of the drying process. Two-phase SOB drying in a fluidized bed proved to be a suitable alternative that provides high-efficiency drying at relatively low air temperatures (T < 150°C). Drying kinetics modeling was developed based on well-known empirical models such as Henderson & Pabiss model and Pages model. An n-order potential model is presented, showing a good accuracy for the description of the experimental drying curves of 2-phase SOB and it was possible to correlate satisfactorily its fitting parameters with the process variables.

Identification of flow faults in continuous reactors by relating linear model parameters and physical magnitudes

A new procedure for predicting dead volume and bypassing in reactors was explored. The method is specific for processes already implemented with a linear reference model. It is based on using a neural network (NN) to obtain relationships between the parameters of the linear model and the dead volume and bypassing. Several experiments with bench scale reactors were carried out and the dead volume and bypassing were found by using classical flow models. By computer simulation we studied the combination of a NN and the linear model of a CSTR with dead volume and bypassing. The NN is a three-layered perception, with sigmoid processing element and back-propagation learning. The input layer receives the parameters of the linear model and the output layer provides the predicted dead volume and bypassing. The accuracy of the trained NN was verified by presenting unseen data to the NN. The prediction errors are less than 15%.

Applications of Neural Networks to pH Control

The neutralization of an acidic waste water was controlled by two NNs. One predicts future pH values, the other manipulates an alkaline current.