Feature Identification for Diagnosing Misalignment under the Influence of Parameter Variation

Misalignment as a result of improper adjustment, heat expansion, and vibration can lead to damage and unexpected downtime of electric motors and their processes. In order to recognize misalignment during processes, or as a simple warning after maintenance, MCSA can be applied. However, previous studies have shown that MCSA fails for load variation and is unable to distinguish faults. At the same time, more sophisticated approaches like deep learning use unknown decision-making processes. Valid features for diagnosing misalignment under the influence of load and motor size variation are unknown. Machine learning algorithms are able to search for valid feature sets. The findings of this paper show that even under load and motor size variation, features for diagnosis can be found. In addition, redundant feature sets with similar results are available and deliver better results than the use of MCSA. The valid features of this study help to implement and to improve technical diagnosis.