A Robust Detection of Broken Bar in Induction Motor using Multilayer Perceptron with Wavelet Transform Feature Extraction

This study uses vibration sensor data in conjunction with Multilayer Perceptron (MLP) deep neural networks and the wavelet transform to offer a reliable technique for identifying broken bar problems in induction motors. The research leverages a publicly available dataset from the IEEE DataPort, focusing on vibration sensor data to achieve high detection accuracy. The robustness of the detection system is evaluated by introducing two kinds of external noise to the testing dataset; Gaussian noise and normal motor vibrations as nearby motor noise, and the model's performance is analyzed across varying levels of noise intensity. Results demonstrate that the MLP with wavelet feature extraction (MLP-W) maintains strong detection capabilities even in noisy environments, promising a practical solution for industrial applications.