Manutenção preditiva em motores elétricos trifásicos utilizando redes neurais artificiais e processamento de sinais

Abstract

This study investigates the application of Artificial Neural Networks (ANN) and signal processing techniques in the predictive maintenance of Three-Phase Electric Motors (TPMs). The objective is to develop an algorithm that, through Motor Current Signature Analysis (MCSA), can identify faults in TPMs, specifically the presence of broken bars in motors operating near nominal torque. The adopted methodology involves collecting measurement data from healthy motors, followed by applying a Sliding Window and the Hamming Window to process the electrical current data. Subsequently, the Fast Fourier Transform (FFT) is used to normalize the resulting vectors in terms of amplitude and frequency. Then, Principal Component Analysis (PCA) is applied to reduce data dimensionality, which is then used as input to a Multilayer Perceptron (MLP) neural network. A case study with a 1 hp TPM was conducted using data obtained from the IEEE Data Port. The ANN, trained and tested with 15 healthy samples and 15 faulty samples (motor with 4 broken rotor bars), achieved an accuracy of over 65% in fault detection. These results indicate that the proposed method is promising for predictive maintenance, showing better performance than previously described approaches in the literature. This work contributes to optimizing the efficiency of TPMs, which are essential for various industrial sectors, reducing energy losses, and increasing the reliability of industrial processes.