Condition Indicator Fusion and Machine Learning-Based Severity Assessment of Tooth Breakage Failures in Spur Gearboxes

Gearboxes play a critical role in the industrial sector due to their efficiency in power transmission. Therefore, detecting a failure in advance is essential. This study aims to determine the severity of a gear tooth failure in a spur gearbox through vibration signal analysis. Under laboratory conditions, a pinion's gear tooth break failure was simulated at different severity levels. Four accelerometers were installed vertically on the gearbox to acquire the vibration signal. Vibration signal features were initially extracted using condition indicators (CIs) in the time, frequency, and time-frequency domains. The CIs from the three domains were fused into a database, and dimensionality was reduced by eliminating highly correlated CIs and those with near-zero variance. The classification accuracy for failure severity level was then determined using the machine learning algorithms Random Forest and k-Nearest Neighbours on the reduced database. Finally, a factorial ANOVA test to asses whether classification accuracy differed significantly across the four sensor positions and two models. The results confirmed that the proposed methodology achieves high classification accuracy in determining the severity of a gear tooth break failure in a spur gearbox. Furthermore, sensor placement on the gearbox has a minimal impact on vibration signal feature extraction.