Application of Data Association and Perceptron Artificial Neural Networks (AR-ANN) in Fault Detection in Dynamic Systems: Gears

Abstract

This work demonstrates a study of identification, classification and grouping of different signals, whose objective is the detection of failures between a pair of gears. Therefore, it is a multidisciplinary work, as it promotes an application of low-cost embedded systems and methodologies of computer science in the area of mechanical engineering. For this to be done, the concept of perceptron artificial neural networks (ANN) associated with the data association rules (AR) theorem belonging to the concept of data-mining was used. This association was developed because it is easy to access and has great potential in identification and classification. We named these different theorems AR-ANN. The result of the application of AR-ANN to the reference and faulty signs was successful, whose classification demonstrated a high rate of correct and in the training phase of the perceptron network, the balance of the adjustment line was obtained, demonstrated by linear regression and weights (variables).