Machine Speed Condition Monitoring using Statistical Time-Domain Features Modeled with Graph

The most important part of speed condition monitoring for industrial machinery is to find the structural changes in dynamic running status. The existing detection methods mainly rely on the time-domain or frequency-domain analysis of the collected operation signals to detect the change. In this paper, the time-domain features and the structural graph similarity are combined to detect the dynamic speed conditions. For the condition signal, we firstly divide it into time-sequential segments. Secondly, extract statistical features from each segment, and modeled as graph to characterize the dynamic behavior of machine speed condition. A common hypothesis testing based on the 30 criterion is finally used to make the change decision. We have investigated the proposed method based on an experimental setup. Experimental results show that compared with the time-domain features, the precision and F score are improved by more than 10%, which shows the effectiveness of our method.