A Comparative Evaluation of Neural Classification Techniques for Identifying Multiple Fault Conditions

The objective of this research is to compare different neural network based classifiers for the accurate and reliable assessment of the status of specific fault conditions in a system with multiple fault conditions present. The proposed strategy utilizes features extracted from vibrational data and employs selforganising maps (SOMs), radial basis function (RBF) and multi-layer perceptron (MLP) networks to model the status of fault conditions, and to discriminate between different fault conditions. Different combinations of vibrational features are evaluated in terms of their ability to support the reliable identification of and discrimination between multiple fault conditions. The results indicate that both SOM and RBF neural classifiers can be trained to reliably identify specific faults in a system with multiple fault conditions present. It is furthermore shown that neural classifiers trained with data reflecting one type of fault mechanism only cannot reliably distinguish between observations with multiple fault conditions present. Key-Words: neural networks, classifiers, self-organising maps, radial basis function, vibrational features, multiple fault conditions