Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
Use Of Artificial Neural Networks In Fault Detection And Diagnosis In Three-Phase Induction Motors
Submission Author:
Guilherme de Oliveira , MG
Co-Authors:
Guilherme de Oliveira, Vinícius Resende Rocha
Presenter: Vinícius Resende Rocha
doi://10.26678/ABCM.COBEM2021.COB2021-1608
Abstract
Owing the current search for computerization, especially supported by the use of artificial intelligence, the industrial maintenance field shows itself to be a fertile environment and with great openness for the use of tools such as artificial neural networks. The application of these resources to equipment as common in the industrial environment as the three-phase induction motors allows an increase in the efficiency of several maintenance KPIs. By monitoring only a few electrical parameters of the motors, neural networks are quite capable of identifying patterns, thus detecting and diagnosing equipment failures. Algorithms capable of simulating the behavior of a three-phase induction motor have been implemented in the Matlab® software, generating the electromagnetic conjugate data, rotor axis speed, stator current and rotor current through a perceptron neural network with 20 artificial neurons in its hidden layers. After the training and validation phases, the developed neural network was tested, in order to define its degree of efficiency in the detection and diagnosis of simulated failures. From the confusion matrices generated for the training, testing and data validation cases, it shows that it can sometimes be difficult for the network in training to differentiate the normal functioning of the engine from its behavior when there are stator or rotor failures, however, 100 % of mechanical or contamination failures were classified correctly. The percentages of correct classifications during the training phase are matched during the validation and test phases. It was demonstrated that at the epoch number 768, the best performance of the ANN was obtained, with an error rate of only 3.1% in the classification of failures. The 96.1% correctness performance during the validation phase was satisfactory compared to the values obtained by other authors. In this way, it is possible to observe that there are simple and relatively accessible ways to bring industrial maintenance activity to lower levels of unexpected failures occurrences.
Keywords
three-phase induction motor, maintenance, Artificial neural networks
