Application of a radial basis function neural network to sensor design

One of the important tasks in sensor design is the development of a model for a sensing phenomena. Artificial neural networks are ideal for such a task because of their capability for representation of the mapping functions describing the processes and phenomena which are mathematically difficult or even intractable. We examined a radial basis function (RBF) neural network for modeling of acoustical properties of colloidal TiO2 slurry. The colloidal slurry is a very complex multiphase medium. The RBF network with a set of local Gaussian functions was trained using the data from the earlier developed physical model of TiO2 slurry. Next the TiO2 neural model was used for a prediction of the TiO2 particle size distribution. The resulting prediction accuracies of the RBF network were 99.8% for the data used in the training process and 88% for the data not used in the training. Compared to other available techniques neural networks can offer an effective and time efficient approach for the modeling of complex materials.