Fusion of vibration based features for gear condition classification.

摘要

Proactive maintenance of drive train components precludes unexpected plant shutdowns. There are different methods to monitor machine conditions. This study focuses on vibration based monitoring.;There are a variety of gear condition indicators available. When applied to a vibration time sequence, different indicators behave differently. Also, because there are so many of them, it is difficult to know which one to choose. This work focuses on how to select the condition indicators that provide maximum discriminatory information and also how to reduce false alarm rates by combining different indicators. Forty-six newly developed gear condition indicators (CIs) were added to the already existing framework of 12 CIs, filling the logical gaps existing among them. CIs were calculated on seeded fault test datasets of an input pinion gear. Thirty-six sets of data were recorded in the test until the failure of the gear. This yielded 58 feature vectors, each with a length of 36. Using engineering judgment, the 36 datasets were divided into 4 classes, starting from the initial gear in good condition to the gear near failure.;Discriminant analysis was applied to the 58 feature vectors to determine a reduce 6 or 7 best features. This set included SLF2, SLF, NB43, FM4, M8A, NBFM43, and NA4. Finally, a simple Baye's classifier was used to combine these features and classify gear condition. The classifier was tested using the training dataset.