Failure mode classification for condition-based maintenance in a bearing ring grinding machine

Abstract Technical failures in machines are major sources of unplanned downtime any production and result reduced efficiency system reliability. Despite the well-established potential Machine Learning techniques condition-based maintenance (CBM), lack access to failure data has limited development a holistic approach address machine-level CBM. This paper presents practical for mode prediction using multiple sensors installed bearing ring grinder process control as well condition monitoring. Bearing rings produced set 7 experimental runs, including 5 frequently occurring critical subsystems. An advanced acquisition setup, implemented CBM grinder, is used capture information about each individual grinding cycle. The dataset pre-processed segmented into cycle stages before time frequency domain feature extraction. A sensor ranking algorithm proposed optimize selection classification installation cost. Random forest models, benchmarked best performing classifiers, trained two-step framework. presence predicted first step type identified second same set. Defining detection improves predictor generalization with classifiers’ performance accuracy $$99\%$$ 99 % on test dataset. presented demonstrates an efficient by selecting crucial resulting cost-effective implementation grinder.