# VFQB: A Novel Deep Learning Model for Rolling Bearing Fault Diagnosis

## TL;DR

This paper introduces a deep learning model for accurately diagnosing faults in rolling bearings, even in noisy environments.

## Contribution

The novel VMD/FFT-Quadratic-BiGRU model improves fault diagnosis by capturing weak features and suppressing noise.

## Key findings

- The model achieved 100% fault diagnosis accuracy on two datasets.
- It outperformed other models by 2.68% and 1.58% in accuracy.
- The model showed superior noise suppression in complex environments.

## Abstract

In rolling bearing fault diagnosis, weak features are often masked by complex environmental conditions, blurring the original fault signals and reducing diagnostic accuracy. To address this issue, we propose the VMD/FFT-Quadratic-BiGRU diagnostic model. First, the original vibration signals are processed with variational mode decomposition (VMD) and fast Fourier transform (FFT) and then stacked as quadratic neural network inputs. Next, a Bidirectional Gated Recurrent Unit (BiGRU) module is introduced to capture the temporal characteristics of the feature signals. An attention mechanism is then applied to assign weights to the hidden layers of the BiGRU network. Finally, fault diagnosis is performed using a fully connected layer and softmax classifier. Experimental results demonstrate that this model significantly enhances the ability to capture weak fault features in complex environments. The fault diagnosis accuracy reaches 100% on both datasets, showing improvements of 2.68% and 1.58% over models without the quadratic network. Additionally, comparisons with other models in noisy environments show that the proposed model exhibits superior noise suppression capabilities, further highlighting its robustness and diagnostic accuracy.