Application of Supervised and Unsupervised Learning Methods to Fault Diagnosis

摘要

Neural networks provide a solution to overcome some drawbacks of the quantitative fault diagnosis because first, they are capable to model linear and non-linear systems off-line. Secondly, they can also learn on-line requiring no a prioriknowledge about the particular system. The neural networks are particularly good for fault diagnosis of systems that have imperfect and/or noisy data. Two basic learning methods and their application to fault diagnosis were studied: supervised andunsupervised learning methods. Two types of neural networks based on supervised learning were considered: the Multi-Layered Percepton networks and the Radial Basis Function networks networks. To solve problem of priori unknown faults, unsupervisedlearning is used. Two types of neural networks based on unsupervised learning investigated: the Kohonen and the Counterpropagation networks. The Radial Basis Function and Counterpropagation network was selected to diagnose faults in a model of anautonomous mobile vehicle as result of analysis of neural networks.