Nonlinear Mapping Lyapunov Exponents-Based KNN Analysis for Fault Classification

Abstract

Rotating machinery are widely used in industry. Machine failures may result in costly downtime. Without effective fault diagnosis, it is impossible to predict the lead-time to failure. Therefore, conducting effective fault detection and identification are desirable and imperative. However, diagnosing faults in rotating machinery is often a labor-intensive and time-consuming practice. This makes conducting effective and efficient fault diagnosis a challenge for technicians and plant maintainers. This paper presents a Lyapunov Exponents-Based K-Nearest Neighbor (LE-Based KNN) analysis method for detecting and classifying rotating machinery fault. To distinguish different machine conditions, Lyapunov exponents (LEs) calculated using nonlinear mapping were selected as features in KNN analysis for classifying the signature of different faults from the machine vibration. The LEs from forty-four vibration recordings served as training dataset for KNN analysis, and the machine health conditions were grouped into four categories. Data from unknown machinery health condition were used as testing data. The results show that proposed LE-Based KNN analysis method can achieve reliable fault classification on operating condition. The LE-Based KNN analysis method has potential improvement in classification with modified KNN and optimized feature selection. This approach can be used to acquire a complete machine vibration profile that may predict occurrence of damage in different parts of machine and help to detect and identify the defective machineries at early stages.