Application of genetic algorithms in the task of choosing inputs for probabilistic neural network classifier of faults of gear-tooth

In this article are presented results of trials of building an application based on probabilistic neural network, used to diagnose damages to the gear wheel teeth in the form of cracks at the base of the tooth. To determine the proper network learning process is necessary to get from the tested object numerous set of input data. Conducted researches are based on data obtained from the identified model of gear working in the drive system, which made it possible to acquire the necessary amount of data. In experiments was tested the usefulness of different sets of descriptors of teeth damages, constructed on the basis of vibratory signals, processed using the Wigner-Ville transform. Often the problem, which makes the proper learning of the neural classifiers impossible is the size of the network structure. Therefore, in further studies was examined the usefulness of genetic algorithms which task is selecting an input data for the artificial neural networks of PNN type.