A Knowledge Discovery Process Extended to Experimental Data for the Identification of Motor Misalignment Patterns

Abstract

The diagnosis of misalignment plays a crucial role in the area of maintenance and repair since misalignment can lead to expensive downtime. To address this issue, several solutions have been developed, and both offline and online approaches are available. However, online strategies using a small number of sensors show a higher false positive rate than other approaches. The problem is a lack of knowledge regarding the interrelations of a fault, disturbances during the diagnosis process, and capable features and feature vectors. Knowledge discovery in database is a framework that allows extracting the missing knowledge. For technical systems, optimal results were achieved by aligning (partially) automated experiments with a data mining strategy, in this case classification. The results yield a greater understanding of the interrelations regarding parallel misalignment, i.e., feature vectors that show good results also with varying load and realistic fault levels. Moreover, the test data confirm a specificity (range 0 to 1) for classification between 0.87 and 1 with the found feature vectors. For angular misalignment, potential vectors were identified, but these need further validation with a modified experiment in future work. For the study, two induction motors with 1.1 kW and 7.5 kW were considered. Furthermore, the findings were compared with additional motors of the same rated power. The findings of this work can help to improve the implementation of sensorless diagnostics on machines and advance the research in this field.