Fault Diagnosis of Rotating Machine Based on Machine Learning with Feature Selection

Fault diagnosis is essential to rotating machines since unpredictable failures can lead to substantial losses in both costs and performance. Although unbalance and misalignment are common types of faults in rotating machinery, they have not been yet considered properly in research. Therefore, this paper proposes a novelty method for fault detection based on the Fast Fourier Transform (FFT) for extracting features from raw vibration signals and the Least Absolute Shrinkage and Selection Operator (Lasso) for selecting the best subset of features. Several machine learning models are then applied for training after feature selection such as Random Forest (RF), Multilayer Perceptron (MLP), and Xtreme Gradient Boosting (XGBoost). Based on the obtained performance results, the diagnosis method of RF achieved an accuracy of 88.63%, which is superior to other machine learning models.