Bringing Bearing Fault Detection to Embedded Systems Using Low-Cost Machine Learning Techniques

Vibration analysis is crucial for predictive maintenance of bearing systems, aiming to reduce costs and prevent failures. This paper investigates signal processing techniques such as FFTs and Hilbert Transforms for fault diagnosis, alongside decision tree models, particularly the J48 algorithm within the WEKA environment, for fault classification. The study emphasizes the integration of signal processing and machine learning for effective fault diagnosis and maintenance. Notably, a classification model with low computational cost has been chosen, facilitating implementation in embedded systems. This model also offers interpretability of outputs, as it is based on decision trees of frequency components obtained through preprocessing.