Transient running stability assessment of industrial rotating machinery based on bag of correlated vibration feature representation

Abstract

The objective of this paper is to monitor and assess the running stability of industrial machinery operating under transient conditions. A new method called bag of correlated feature representation (BOCFR) is presented in this paper, which aims to extract structural information hidden in the data. At first, the bag-of-words (BoW) model is employed to cluster the collected training samples. For a new observed signal, a correlogram is subsequently constructed with BoW to reflect its dynamic characteristic. Entropy is then calculated, with the aim of periodically analyzing the dynamic characteristics of machine running status over time. Together with BOCFR, a unified framework by means of hypothesis testing is finally proposed, which is evaluated based on simulation scenarios and real-engineering applications: a comparison of results with those of typical techniques demonstrates that the performance of the proposed method is promising.