Diagnosis of Localized Bearing Defects using Texture Analysis of Vibration Envelope Signals and Machine Learning

Bearings are important because they play a critical role in the operation of many machines and devices by reducing friction and enabling smooth motion between two moving parts. They are commonly used in rotating machinery, such as electric motors, turbines, pumps, and gearboxes, where they support and guide the rotating shafts. The primary cause of failure in wind turbines and induction motors is bearing faults, resulting in significant downtime and economic losses. Detecting these faults accurately and promptly is crucial to preventing unexpected shutdowns. Most of the existing methods for bearing fault detection rely on envelope analysis, which involves filtering and demodulating the raw signal, and Fourier analysis of the envelope signal to identify defect frequencies. Recent research has focused on improving the visibility of these frequencies by selecting an optimal band for filtering, using sub-band analysis and spectral kurtosis, which is a complex process and can be time-consuming. This paper presents a new approach to envelope signal analysis using texture analysis of the envelope signals that are projected as 2D grayscale images where each fault generates a unique texture. These textures are encoded using the local binary pattern operator and then used for fault classification. The method is tested on a publicly available seeded fault dataset.