Vibration Signal Analysis Towards Early Detection of Machine Faults

Abstract

Intelligent fault diagnosis of rotating machines is an important task towards the reliability and safety of modern industrial systems. Complex and variable operating conditions of machinery makes early detection of faults and their patterns more challenging. In this paper, we proposed a vibration signal-based pattern analysis framework for the effective detection and classification of machine faults. 3-axis vibration signals were first acquired from faulty and non-faulty machines using a sensitive accelerometer. Vibrations signals were independently preprocessed through empirical mode decomposition (EMD) to capture the most discriminating signal components while discarding the redundant information. Subsequently, significant time and frequency domain features were extracted from each channel of the preprocessed signal and fused to form a strong feature representation of each class. Features were selected from a pool of time, frequency, and statistical domain after extensive experimentation. The fused feature vector was employed to train and test support vector machines (SVM) classifier through 10-fold cross-validation. Comparative analysis of base classifier (SVM) with K-nearest neighbor (KNN) and Decision Tree (DT) is also presented. The proposed method yields the best results in terms of accuracy of 99%, sensitivity of 100%, and specificity of 99%.