Classification of chemical systems using acoustic emission and neural networks

A novel acoustic wave sensor capable of classification of chemical reactions has been designed, fabricated and tested. The principle of sensor operation is based on the acoustic emission phenomena. The proposed acoustic emission chemical sensor consists of four sections: the measurement cell with ultrasonic transducers operating in the frequency range from 90 kHz to 2 MHz, the frequency domain signal detection unit, the signal processing unit based on a neural network and a computer controlled data acquisition system. A Probabilistic Neural Network (PNN) has been implemented for classification of chemical systems. The centers of the Gaussian nodes which construct the PNN are the individual points of the training data. The Gaussian nodes are organized with respect to the class information. A test pattern is applied to the PNN, and the Gaussians are summed for each class. The class with the greatest probability, or largest sum, is the network output. The width of the Gaussians, sigma, is optimized to give the best classification on the training data. Preprocessing techniques have been designed to extract features of the data for use as an input of the neural network.