A Novel Approach to Vibro-Acoustic Fault Diagnosis Using Intrinsic Multidimensional Gnome Entropy Analysis and a Convolutional Neural Network

Due to the complex practical working conditions, fault diagnosis is very challenging in rotary machinery. Rotary machinery is often used in places with a high temperature, strong corrosion, or where acceleration sensors are hard to install. Thus, acoustical signal processing becomes a powerful tool for diagnosing faults in rotary machinery. This paper proposes a novel strategy called intrinsic multidimensional Gnome entropy (IMGE) analysis that can extract the quantitative features of vibro-acoustic signals. The method consists of signal analysis using variational mode decomposition (VMD), and feature extraction using the proposed mean Gnome entropy (MGE). Multidimensional MGE solves the issue of negative computation values in the original algorithm and improves the accuracy and stability of entropy in multiple dimensions. VMD is used to decompose vibro-acoustic signals into a series of band-limited intrinsic mode functions (BLIMFs) denoting the dynamical characteristics of signals. Multidimensional MGE values of BLIMFs are extracted as quantitative features and inputted into a convolutional neural network (CNN) to diagnose faults in rotary machinery. The improved performance of the proposed framework under the tested experimental conditions is verified by theoretical derivation, numerical simulations, and research on practical applications.