Gear crack level classification based on multinomial logit model and cumulative link model

In order to avoid machine related catastrophes, the early detection of cracks is in urgent demand. Sensors are put into the rotating parts of machine and vibration signal data are collected to diagnose machine health. This paper proposes a comprehensive method to look into the development of damage with multinomial logit model (MLM) and cumulative link model (CLM). We first select features according to analysis of variance (ANOVA), and then compare the MLM, CLM method with weighted k-nearest neighbor method (WKNN) - a black box machine learning algorithm and we conclude that these methods have their pros and cons in the diagnosis of faults.