Técnicas de caracterização de excitações em máquinas rotativas

Abstract

This work presents a contribution to the study and characterization of excitements forces applied in rotative machine, using neural netwoks starting from the measured vibration sign in the equipment. The developed methodology is used to classify the excitement among four states of operation conditions: normal operation, umbalance excitement force, asynchronous excitement force and magnetic excitement force. For the each excitement class it is possible to determine its application point and its amplitude and frequency characteristics. Applying a technique statistics data compress of the measured signs, it was possible to training with success neural networks with smaller number of neurons, consequently with smaller computational cost. The efficiency and robustness of the architectures proposals, of neural networks, they were appraised for different levels of data compress and of addictive noises using numeric simulation of a vibratory model of systems of three degrees of freedom. The methodology was validated in a experimental apparatus that represents rotative machine whit a flexible rotor.