Feature selection algorithm based on density and distance for fault diagnosis applied to a roll-to-roll manufacturing system

Abstract Roll-to-roll systems that include rotary components such as driven rolls and idle rollers have significant potential for application in fabrication of flexible functional devices. They are inexpensive, mass producible, environmentally friendly; however, even minor defects their component bearings can render them susceptible to severe damage, which necessitates accurate diagnoses bearing quality. The main steps machine learning fault diagnosis feature extraction selection. In the case high-dimensional data, critical study is required identify best combination proper diagnosis. Thus, this aims develop a method extracts characteristics from measured signal qualify according Mahalanobis distances differences density between normal faulty data groups. Features extracted vibration collected industry-scale roll-to-roll CWRU were trained with principal analysis, other modern selection techniques, proposed algorithm-based eight classifiers. Compared existing algorithm, accuracy increased by up 9.24%, training time decreased 34.46%, number features obtain maximum 59.92%. algorithm provides an effective time-efficient approach improve components.