Investigation into Bearing Fault Classification using Various Feature Set Combinations in KNN

In many rotating types of machinery, rolling element bearings are considered "crucial", as rotating components cause the majority of failures, with proper performance monitoring; imminent failures can be identified and corrected. At the initial phase, there may exist many distinct features embedded in the data that may be able to identify a bearing fault which are non-traceable by using the physical models. Due to which, Machine Learning (ML) algorithms are currently the growing area of research for diagnostic and prognostic purposes. This work presents a study on the fault classification accuracy by K-Nearest Neighbor (KNN) machine learning classifier wherein the effect of different Time domain, Frequency domain and a combination of Time and Frequency domain features individually and in combinations have been investigated. The classification accuracy of Fine, Medium, and Coarse KNN have been examined and from the results obtained, Fine KNN has been used to further study the classification accuracy for various feature set combinations. Furthermore, features have been ranked using One-way ANOVA and their classification accuracies obtained using those top ranked features and their combinations have been investigated.