An Original feature retention deconvolution algorithm and its application to bearing fault diagnosis

Abstract

Abstract Traditional blind deconvolution (BD) algorithms perform well in estimating the repetition rate of impulses within signals; however, they fall short in preserving the original features of the signal. In engineering applications, partic-ularly for cyclic impulse signals, maintaining signal fidelity is as crucial as accurately estimating impulse counts, making pure impulse count estimation insufficient for practical needs. To address this limitation, we propose a novel deconvolution algorithm—Maximum Correlation Pearson Fidelity Coefficient Deconvolution (MCPSFD). This method constructs an objective function based on two key metrics: the Correlation Pearson Coefficient (CPC), which quantifies the periodicity of impulses, and the Signal Fidelity Coefficient (SFC), which measures the similar-ity between the original and recovered signals. By combining CPC and SFC, we introduce a new objective function, termed the Correlated Pearson Fidelity Factor (CPSF), which simultaneously considers both the number of impulses and the original characteristics of the filtered signal without introducing redundant parameters. The MCPSFD algo-rithm is derived by maximizing the CPSF function. Extensive experiments on simulated and measured bearing sig-nals demonstrate that the proposed method significantly outperforms existing deconvolution algorithms in recover-ing periodic impulses and minimizing signal distortion.