Work

Fault Diagnosis of Ball Bearings Using Fast Continuous Wavelet Transform and XGBoost on a Multi-Condition Acoustic Dataset

Year: 2026
Type: article
Abstract: Reliable fault diagnosis of rotating machinery, particularly ball bearings, is critical to ensuring operational safety and minimizing downtime. While acoustic signals provide rich diagnostic information, Artificial Intelligence (AI) models often suffer from overly optimistic metrics due to data leakage during preprocessing and splitting stages. This study presents a robust fault diagnosis framework with minimized leakage risk, using a multi-condition acoustic dataset. To completely eliminate leakage, a strict evaluation pipeline is implemented: (i) window-wise normalization is applied to each raw signal window prior to the Fast Continuous Wavelet Transform (fCWT) to prevent information sharing between windows, and (ii) data partitioning is strictly managed via Group Shuffle Split (GSS) and Stratified Group K-Fold (SGKF) cross-validation, ensuring that signals from the same recording do not span across training and test splits. Time–frequency features extracted directly from fCWT coefficients are processed using Extreme Gradient Boosting (XGBoost), which serves a dual purpose as an embedded feature selection mechanism and the core classifier. The framework achieves realistic and generalizable performance under distinct operational speeds (800, 1000, and 1200 RPM) and load conditions (0%, 15%, and 30%), yielding an F-score of 99.81% under GSS and a mean F-score of 99.35% across SGKF cross-validation, providing a trustworthy benchmark for health monitoring.
Institution: Fırat University
Language: English
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7236031560
FWCI:
Cites: 53
Cited by:
Related to: 10