Abstract:Strong background noise, multiple interferences, and complex transmission paths contaminate weak early fault signatures of vibration signals from rotating machineries, hampering timely and effective fault detection, thus affecting operational safety. Although early fault signal enhancement methods have attracted attention, the traditional blind deconvolution (TBD) method remains sensitive to noise and impulsive interferences despite extensive research conducted. It often falls into local optima during iterative optimization, limiting its flexibility and reliability in enhancing early fault pulse components. Additionally, its heavy reliance on priori knowledge of repetitive pulse periods restricts its practical application. To address aforementioned issues, a multi-objective optimal deep deconvolution (MODD) method is introduced. A dual-branch collaborative deep convolutional network is constructed by the MODD method, which jointly optimizes the dual-branch collaborative deep convolutional network by employing the correlation kurtosis and the noise energy multi-objective loss function as the joint optimization objective. Furthermore, the MODD method precisely identified the fault periods of rotating machinery vibration signals during the optimization iterations by utilizing the autocorrelation function. By incorporating the proportion factor of fault pulse components, it determines the location of early fault pulse components. On this basis, the MODD method constructs vectors representing repetitive pulses and noise energy, reflecting the repetitive pulse and noise components that are closely associated with early faults. A multi-objective loss function is established to enhance pulse components and suppress interference, optimizing fault-related signals while mitigating the effects of noise and random pulses. The effectiveness and generalization ability of the proposed MODD method are validated through experiments on rolling bearing fault signal collected from aero-engine accessory gearbox, and an enhanced performance is obtained when compared with traditional TBD methods.

Convolutional Sparse Learning for Blind Deconvolution and Application on Impulsive Feature Detection.

Robust Sparse Threshold Optimization for Impulsive Blind Deconvolution

Blind Deconvolution Using an Improved L0 Sparse Representation

Convolutional plug-and-play sparse optimization for impulsive blind deconvolution

Feature Extraction Of Impulse Faults For Vibration Signals Based On Sparse Non-Negative Tensor Factorization

Nonnegative Bounded Convolutional Sparse Learning Method for Envelope Feature Deconvolution.

Nonlinear Blind Deconvolution Based on Generalized Normalized l p / l q Norm for Early Fault Detection

Nonnegative bounded convolutional sparsity learning algorithm for envelope blind deconvolution

An Adaptive Multi-D-Norm-Driven Sparse Unfolding Deconvolutional Network for Bearing Fault Diagnosis

A novel bearing intelligent fault diagnosis method based on spectrum sparse deep deconvolution

Low-rank enhanced convolutional sparse feature detection for accurate diagnosis of gearbox faults

Investigation on the Kurtosis Filter and the Derivation of Convolutional Sparse Filter for Impulsive Signature Enhancement

Mechanical Vibration Signal Separation Using Blind De-Convolution Technology

Maximum correntropy criterion-based blind deconvolution and its application for bearing fault detection

Manifold Sensing-Based Convolution Sparse Self-Learning for Defective Bearing Morphological Feature Extraction

A simplified convolutional sparse filter for impulsive signature enhancement and its application to the prognostic of rotating machinery

Blind Deconvolution Technique Based on Improved Correlated Generalized Lp / Lq Norm for Extracting Repetitive Transient Feature

Resilient fast convolutional sparse filtering with convergence directivity for the bearing fault diagnosis

Multi-objective Optimal Deep Deconvolution and Its Application to Early Fault Signal Enhancement of Rotating Machineries

Blind Deconvolution Technique Based on Improved Correlated Generalized <i>Lp</i>/<i>Lq</i> Norm for Extracting Repetitive Transient Feature

Iterative Morlet Wavelet with SOSO Boosting Strategy for Impulsive Feature Extraction