Optimal weighted multi-scale entropy-energy ratio feature for machine fault diagnosis

Abstract

Identifying the sensitive characteristics of mechanical equipment components is crucial for effective fault diagnosis. However, focusing solely on a specific feature at a single time scale fails to comprehensively capture the device’s operational state. Inspired by the concept of multi-scale analysis and recognizing the complementary strengths of permutation entropy (PE) and root mean square (RMS) in fault characterization, we propose a novel feature called the Optimal Weighted Multi-Scale Entropy-Energy Ratio (OWMEER). This feature aims to enhance fault characterization by optimally combining the strengths of PE and RMS, thereby providing a more comprehensive assessment of the device’s condition. The effectiveness and superiority of OWMEER in fault characterization have been validated through experimental data, including both public and self-test datasets, when combined with the commonly used pattern recognition methods such as random forest (RF) and support vector machine (SVM). The results demonstrate that using OWMEER as a fault feature not only yields better results than using the original features RMS and PE, but also maintains strong diagnostic performance across different classifiers and datasets.