A Functional Deep Learning Framework for Rotary Machines Fault Detection

Read the full article See related articles

Listed in

This article is not in any list yet, why not save it to one of your lists.
Log in to save this article

Abstract

Accurate and robust fault diagnosis of rotary machines is essential for maintaining the safety and reliability of industrial systems. Although deep learning methods have achieved remarkable success in this domain, their performance tends to decline in the presence of noise. In this study, a functional deep learning framework is introduced, in which raw vibration signals are transformed into smooth functional representations prior to being processed by deep learning architectures. By incorporating functional data analysis (FDA), the noise robustness and generalization ability of standard models are enhanced. Functional adaptations of several well-established one-dimensional convolutional networks are implemented and evaluated on a benchmark dataset under varying signalto-noise ratios (SNRs). The experimental results demonstrate that functional representations consistently improve diagnostic accuracy, particularly in lowSNR environments. These findings underscore the effectiveness of functional techniques in enhancing deep learning-based fault diagnosis of rotary machines operating under noisy conditions.

Article activity feed