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

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Impulsive blind deconvolution (IBD) is crucial for bearing fault diagnosis, relying on objective functions and transfer functions.
• Existing IBD methods struggle to simultaneously preserve waveform impulsiveness and periodicity cyclostationarity.
• The single convolution operation in IBD is inadequate for complex transmission paths involving multiple linear and nonlinear units.

Purpose of the Study:

• To propose an adaptive multi-D-norm-driven sparse unfolding deconvolution network (AMD-SUDN).
• To address limitations of current IBD methods in retaining signal impulsiveness and cyclostationarity.
• To enhance the accuracy of detecting impulsive features in bearing faults.

Main Methods:

• Developed a novel target vector construction using MaxPooling period modulation intensity (MPMI) for simultaneous impulsiveness and cyclostationarity.
• Designed objective functions incorporating a multi-D-norm driven approach based on the constructed target vector.
• Unfolded an iterative soft threshold algorithm (ISTA) for convolutional sparse learning (CSL) into a deconvolution network (AMD-SUDN).

Main Results:

• Numerical simulations validated the algorithm's performance and optimal hyperparameter configurations.
• The AMD-SUDN successfully detected impulsive features characteristic of bearing faults.
• Comparative analyses demonstrated superior deconvolution accuracy of AMD-SUDN over existing state-of-the-art IBD methods.

Conclusions:

• The proposed AMD-SUDN effectively overcomes limitations of traditional IBD techniques.
• The method achieves enhanced deconvolution accuracy for bearing fault diagnosis.
• AMD-SUDN offers a promising advancement for condition monitoring and fault detection in rotating machinery.