Gearbox Broken Tooth Severity Classification using EMD of Acoustic Emission Signals

Fault severity classification is a critical task necessary for optimal predictive maintenance to reduce costs and avoid catastrophic accidents in the industry. In this research, we propose a methodology for broken tooth severity classification in a gearbox using digital signal processing techniques of acoustic emission signals. The method uses empirical mode decomposition of the signal and extraction of time-domain features from a set of Intrinsic Mode Functions. The extracted features are fed to random forest and linear discriminant analysis models for attaining the classification of nine different severity conditions. The method provides classification accuracies higher than 90% with both machine learning models.