Enhancing turbine gear fault monitoring through the integration of envelope analysis and recurrent neural networks

Said Djaballah, M'hamed Beriache, Abdelmoumene Hechifa, Kamel Meftah and Abdelhak Belahcene

Abstract

Gearbox is an integral part of wind turbine (WT) design and timely fault detection can reduce unexpected downtime and maintenance costs. This study presents a method to detect turbine gear faults that combines Envelope analysis combines with Gated Recurrent Units (GRUs). Envelope analysis extracts high-frequency fault features from vibration signals, while GRUs excel in recognizing temporal patterns, making this combination particularly powerful for early fault detection in gears. This method aims to enhance diagnostic accuracy, offering significant advantages over traditional methods that rely primarily on human inspection and basic signal processing. Its ability to detect and localize issues early ensures a direct and impactful contribution to optimizing maintenance strategies. The results, based on the analysis of signals from a planetary gearbox, demonstrate a marked improvement in diagnostic capabilities.

Key Words

gated recurrent units; gear faults; envelope analysis; recurrent neural networks; vibration signals

Address

Said Djaballah and Abdelhak Belahcene: Department of Mechanical Engineering, Faculty of Technology, University of Chlef, Algeria M'hamed Beriache: Rheology and Mechanics Laboratory, Faculty of Technology, University of Chlef, Algeria Abdelmoumene Hechifa: Department of Mechanical Engineering, Faculty of Technology, University of Skikda, Algeria Kamel Meftah: LGEM, Department of Mechanical Engineering, Faculty of Technology, University of Batna 2, Algeria

← Back to Volume 12, No. 2