CLASSIFICATION OF AUTOREGRESSIVE SPECTRAL ESTIMATED SIGNAL PATTERNS USING AN ADAPTIVE RESONANCE THEORY NEURAL-NETWORK

Machine condition monitoring and fault detection has been an important issue for manufacturing practitioners and researchers around the world, as it impacts production efficiency and effectiveness as well as the morale of the production crew profoundly. This paper examines the use of a relatively new technology, Adaptive Resonance Theory (ART), to assess the machine condition through vibration signals. The vibration signal is first compressed with an Autoregressive (AR) technique in order to reduce the amount of information which the ART neural network is to deal with. The theoretical foundation of the fault classification system is discussed, followed by a brief case study.