Three-Stage Method for Rotating Machine Health Condition Monitoring Using Vibration Signals

This paper proposes a new three-stage method for rotating machines health condition monitoring. In the first stage of the proposed method, Multiple Measurement Vectors Compressive Sampling (MMV-CS) is used to obtain compressively-sampled signals from the acquired raw vibration signals. In the second stage, a process combining Geodesic Minimal Spanning Tree (GMST), Stochastic Proximity Embedding (SPE), and Neighbourhood Component Analysis (NCA) is used to estimate and further reduce the dimensionality of the compressively-sampled signals. In the third stage, with these reduced features, multi-class Support Vector Machine (SVM) classifier is used to classify machine health conditions. Experiments on a roller element bearing fault detection and classification task based on vibration signals are used to verify the efficiency of the proposed method. Results show that the proposed method with fewer features achieved high classification accuracy of bearings health conditions and outperformed recently published results.