DETECTION OF IMPERFECTIONS IN ROTATING MACHINES THROUGH ARTIFICIAL NEURAL NETWORKS
Jorge Nei Brito, Davison Fagundes Portes, Francianne Tavares, Geraldo Ronivan Pinto · 2005
The artificial neural networks (ANNs) are mathematical and computational models inspired in the knowledge of the neurosciences. The ANNs is composed of nonlinear elements of parallel processing and is characterized for the capacity of learning through examples. In this work a contribution of the study and characterization of forces of excitement in rotating machines using ANNs trained with experimentall signals of vibration is shown. The developed methodology is used to classify the excitement for four levels of unbalancing beyond the normal condition of functioning. Through a selective filter it is possible to reduce the number of parameters capable to represent the signals used for training of the ANNs. To the results of training through batch method and standard-to-standard method and of qualification for different architectures of networks is shown. The evaluation of sensor's efficiency related with the point of vibration signals acquired is shown too. In this way the goal is to identify which sensors presents the percentage greater of rightness reducing the number of collections.