Parameter-adaptive CYCBD β blind deconvolution for bearing fault diagnosis of a reduction gearbox in mixer trucks

Abstract

A novel deconvolution technique, termed CYCBD _β blind deconvolution (CYCBD _β ), has recently been proposed. Compared to CYCBD and other blind deconvolution (BD) methods, CYCBD _β demonstrates superior performance in extracting fault-induced weak cyclic impulses under both Gaussian and non-Gaussian noise conditions (e.g., random impulsive noise). However, a major challenge in applying CYCBD _β to real-world scenarios is the appropriate selection of several critical parameters, particularly the target cyclic frequency and filter length. This is primarily because the effectiveness of CYCBD _β is highly dependent on the accuracy of the provided period information and filter size. To overcome this limitation, this study proposes a parameter-adaptive version of CYCBD _β , referred to as PA-CYCBD _β . The proposed method integrates an effective tool, an envelope harmonic product spectrum, specifically designed to accurately estimate the true cyclic frequency. Following this, a newly developed optimization algorithm, the escape optimization algorithm, is employed to determine the optimal filter length. Compared to the original CYCBD _β , PA-CYCBD _β can detect weak impulse features embedded in raw vibration signals without requiring prior knowledge of the periodicity and filter length. The effectiveness and advantages of PA-CYCBD _β are validated through both simulated signals and experimental data collected from a reduction gearbox on a concrete mixer truck test platform.