Using model updating technique to train neural network for fault detection

• Autor(a) principal: Lopes, Vicente [UNESP]
• Data de Publicação: 1997
• Outros Autores: Pereira, João Antonio [UNESP], Weber, Hans Ingo
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1115/DETC97/VIB-4233; http://hdl.handle.net/11449/228909
• topic: Fault classification; Model updating; Neural network; Predictive maintenance

Vibration monitoring and fault detection of components in manufacturing plants involve a detailed analysis of a collection of vibration data in order to establish a correlation among changes of the measured data and the corresponding fault. This work presents an alternative proposal which intent is to exploit the capability of model updating techniques associated to neural networks to reduce the amount of measured data. The updating procedure supplies a reliable model that permits to simulate any damage condition, which allows to establish a direct correlation between the deviation of the response and the corresponding fault. The learning of the net is performed using a compressed spectrum signal created for each specific type of fault. Different fault conditions for a frame structure are evaluated using simulated data and finally, the capability of the proposal is demonstrated using experimental data.