Machine learning-based fault diagnosis of rotating machinery using acoustic data

Abstract

The industrial advancement has promoted the development of machine learning based intelligent fault diagnosis methods for condition-based maintenance. Various condition-monitoring techniques can be used. However, the most reliable approaches require complex and high-cost data acquisition setups. This led to the use of acous-tic signals for fault diagnosis in this study. The study presents a machine-learning fault classificatio napproac htha tleverage sfeature sextracte dfro mth edecomposed acoustic signals using Empirical Mode Decomposition (EMD) and Maximal Overlap Discrete Wavelet Packet Transform (MODWPT) decomposition methods. The classificatio ni sperforme dusin galgorithm sconsitin go fSuppor tVecto rMachines (SVM), K-Nearest Neighbors (KNN), Decision Trees, and Ensemble Bag. These machine-learning algorithms have been tested through different experiments to evaluate the proposed approach on two datasets, MAFAULDA Machinery Fault and the Air Compressor datasets. The results revealed that SVM exhibited superior accuracy and out performed other classifiers in most evaluation metrics. Also ,it demonstrated robustness in noisy environments, and exhibited the fastest prediction time. Decision tree demonstrated that it is the most storage-efficie ntmodel.