A novel rotor-stator rub-impact fault and locations recognition method

Rub-impact fault is a kind of typical secondary fault encountered by large-sized rotation machinery. When it happens, the fault information contained is often weak, complicated and compound. This makes it difficult to correctly identify a rub-impact fault and affected part. Signals captured by sensors vary greatly in terms of energy, stability and complexity before and after a rub-impact fault occurs and when affected part is different. Graph indexes are sensitive to the variation of energy, stability and complexity in graph signal. Based on this, the paper has raised a method based on graph indexes to identify rub-impact fault and its locations and graph indexes are used to describe different running status of equipment. Firstly, to reduce the influence of noise components, vibration signals are given an auto-correlation analysis. Secondly, each auto-correlation function is converted to graph signals and five graph indexes of each graph signal are calculated. Thirdly, when running state of equipment is the same, an analysis is given to the consistency of graph indexes of signals from the same sensor, as well as the difference of graph indexes of each signal from different sensors. Fourthly, when running state of equipment is different, the difference of graph indexes of signals from the same sensor is analyzed. Fifthly, an analysis is given to the effectiveness and sensibility of graph indexes of signals from different sensors to running state. Finally, by the excellent consistency and difference shown by graph indexes, it is treated as a feature vector and according to two classical classification algorithms, rub-impact fault and location is identified. The result indicates that the mean identification rate of proposed method for rub-impact fault and locations can reach to 96%; graph indexes can serve as fault feature index to effectively represent a rub-impact fault and locations and has very excellent robustness.