Gear Anomaly Detection Using a Matrix Profile Index: Fixed-Axis Gearbox with Cracked Gear: An Application Using Euclidean Distances

Author(s):

Arnon Hurwitz and Adrian Hood

Abstract:

Our purpose is to describe a new anomaly detection method for a cracked spur gear running in a fixed-axis gearbox. Our basic assumption is that the vibration waveform taken from one record of a gear that is running free of anomalies matches best in “Euclidean Distance” to either the record that went before it or the record after. In the present study of six gears, we compare detection outcomes using the new approach with standard vibration-based gearbox Condition Indicator (CI) methods and with deep-learning artificial intelligence (AI) methods exercised on the same data sets. Our method is shown to compare favorably to these alternatives. We describe a new, data-driven anomaly detection method for a cracked spur gear running in a fixed-axis gearbox. Anomaly detection is the first step of a health and usage monitoring system for machinery. The “newness” of the proposed technique resides in the application of a “Matrix Profile Index” to flag anomalous behavior. In the present study, we compare detection outcomes using the MPI approach with standard vibration-based gearbox CI methods and with deep-learning AI methods exercised on the same data sets.