Minimum envelope-derivative entropy deconvolution: A novel blind deconvolution method for rolling bearing fault diagnosis

Abstract

Extracting weak repetitive transient features from vibration signals under heavy noise and random discrete impulses remains a critical challenge in rolling bearing condition monitoring. To address the limitations of traditional blind deconvolution methods, this paper proposes a novel technique termed minimum envelope-derivative (ED) entropy deconvolution (MEDED). MEDED introduces ED-entropy as a new objective function, which effectively characterizes the cyclostationarity and sparsity of fault transients while maintaining robustness against random impulses. An iterative algorithm based on normalized gradient descent is derived to minimize the ED-entropy of the filtered signal, thereby recovering periodic fault impulses. The method is rigorously validated using simulated signals and experimental datasets. Comparative studies against state-of-the-art techniques confirm that MEDED significantly outperforms existing methods in noise suppression and the extraction of clear fault characteristic frequencies. Consequently, MEDED presents a robust and effective alternative for bearing fault diagnosis in practical engineering applications.