Supervised Machine Learning Based Approach for Early Fault Detection in Polymer Gears Using Vibration Signals

摘要

The comprehensive mechanical features of polymer gears have demonstrated their potential for power transmission. The primary concern with expanding their power transmission applications is developing a reliable fault detection technique. In this manuscript, an early fault detection method of polymer gear is proposed using supervised machine learning with Hjorth parameters (HP). In this work, HP is used along with traditional statistical features to improve the polymer gear fault detection accuracy from vibration signals in the early stage. Vibration signals are acquired experimentally at a different class of polymer gear faults, namely healthy, slight, moderate, and severe pitting. The HP and SF vectors are extracted from the optimum intrinsic mode function of empirical mode decomposition and empirical wavelet transform (EWT). After that, faults are classified by three different classification algorithms, namely support vector machine, linear discriminant analysis, and K-nearest neighbor (KNN). Evaluating the result of these three classifiers with three sets of feature vectors (SF, HP and a combination of SF and HP) and comparing all the results with raw signal data. The result showed that the maximum accuracy, i.e., 99.3%, is achieved by KNN classifier with EWT method for signal decomposition using a combination of HP and SF vector.