Fault Detection in Ball Bearings

TL;DR

This paper investigates the use of vibration image preprocessing combined with convolutional neural networks for fault detection in ball bearings, demonstrating robustness on larger datasets and hyperparameter analysis.

Contribution

It introduces an enhanced vibration image preprocessing method for CNN-based fault detection, validated on larger datasets and with hyperparameter exploration.

Findings

Vibration images improve fault detection accuracy.

The method is robust across larger datasets.

Hyperparameter tuning affects detection performance.

Abstract

Ball bearing joints are a critical component in all rotating machinery, and detecting and locating faults in these joints is a significant problem in industry and research. Intelligent fault detection (IFD) is the process of applying machine learning and other statistical methods to monitor the health states of machines. This paper explores the construction of vibration images, a preprocessing technique that has been previously used to train convolutional neural networks for ball bearing joint IFD. The main results demonstrate the robustness of this technique by applying it to a larger dataset than previously used and exploring the hyperparameters used in constructing the vibration images.