Optimal Relative Entropy Driven Envelope Descriptors for Robust Fault Diagnosis of Travel Gearbox in Excavator

This research presents a novel fault diagnosis methodology for excavator travel gearboxes, integrating envelope analysis and a multi order probabilistic approach (MOPA) algorithm. Vibration signals obtained in industrial field conditions often suffer from noise and speed fluctuations, complicating their analysis. This study addresses this challenge by quantifying uncertain results using relative entropy, a probabilistic distance metric, leading to robust diagnostic outcomes that are resilient to varying driving conditions. A critical aspect of this study is its utilization of the observed smearing phenomenon in the envelope spectrum distribution of nonstationary signals, which is inherently related to the speed distribution. This observation forms the basis of a novel fault diagnosis approach, utilizing features derived from this phenomenon. To mitigate the parameter dependency of conventional MOPA and band-pass frequency region for desirable Hilbert transforms, both methodologies are integrated, and parameters are automatically optimized using a fault diagnosis-oriented optimization approach. Comparison with a non-optimized method and resampling based order analysis validates the superior performance of the proposed optimization for fault diagnosis. The proposed Relative Entropy driven Envelope Descriptor (REED) feature is applied to a support vector machine and demonstrates excellent fault diagnosis performance as evidenced by coefficient analysis results.