Vibration feature extraction via graph-modeled SVD for running stability assessment of industrial machine

Abstract

Detection of structural changes in machine running state has been a long-standing problem in the monitoring and prognosis of industrial machinery. This paper presents a new method for this purpose. The singular values are obtained from collected vibration signals based on singular value decomposition (SVD). The graph is constructed from the resulting singular value series. Anomaly score is calculated between two graphs. Change decision is made by testing a null hypothesis testing based on the anomaly scores. The proposed method is investigated based on an experimental system setup and the results demonstrate significant improvements over some benchmark methods reported in the literature.