•  
  •  
 

Date of Acceptance

February 2026

Date of Submission

November 2025

Abstract

When machines break down unexpectedly, it costs factories a lot of money and can even put workers at risk. That is why vibration analysis has become such a big deal in predictive maintenance. By listening to how a machine vibrates, engineers can often tell something is wrong before the whole thing falls apart. This paper looks at how we have gotten better at this over the years. We started with fairly basic signal processing, things like Fourier transforms that have been around for ages, and now we are seeing some really interesting work with neural networks and deep learning. I went through a lot of the published research on this topic, covering everything from how we collect vibration data to how we process it and classify different fault types. What seems to work best right now is combining wavelet analysis with deep learning models. Some researchers are reporting accuracy rates over 97% for detecting bearing faults, which is quite impressive. But there are still problems to solve. Getting enough labeled training data is hard, and models trained in the lab do not always work well on real factory floors. This review tries to lay out where we are now and where things might be headed.

Included in

Manufacturing Commons

Share

COinS