Level: Advanced
Gwang-gi Lee, Managing Director, Haneol Solution
In order to build a real-time abnormality detection monitoring system for roller bearings, the acceleration signal of the roller bearing is preprocessed in 0.5 second increments and then various statistical indicators are applied to the preprocessed acceleration signal to determine the NOR (normal) and ORF (outer race) of the roller bearing. After defining the patterns of (inner race fault) and IRF (inner race fault), we learn the patterns using machine learning techniques to build a system that monitors abnormality detection in real time. By applying JMP's statistical methodology, Skewness and Kurtosis were selected as statistical indicators that recognize NOR (normal), ORF (outer race fault), and IRF (inner race fault) patterns of roller bearing acceleration signals, and pattern recognition is provided by JMP. An anomaly detection model was built by applying machine learning methodologies such as ANN (Artificial Neural Network), SVM (Support Vector Machines), and logistic regression, NL (Nominal Logistics).
[video]
